What’s New

3.14.1 (2026-11-18)

Bug fixes

  • CD FLAC frames no longer overflow their buffer (#69922). Recompressing a CD CHD that stores one frame per hunk could fail with FLAC frame too large: 2419 bytes (buffer 2416) when a valid Fixed subframe needed slightly more metadata than the 64 bytes reserved per frame. The CD-FLAC and laserdisc A/V audio encoders now reserve 512 bytes of headroom per FLAC frame.
  • Empty CLI output paths report a usage error (#69897). CHDSharp extractcd -o "" now prints the missing-argument message and help text instead of throwing an ArgumentException from Directory.CreateDirectory.
  • MDS Mode 2 Form 1 and Form 2 tracks convert (#69872). Cooked 2048-byte Mode 2 Form 1 tracks are described as MODE2/2048 and 2324-byte Form 2 tracks as MODE2/2324, both understood by chdman. These descriptors no longer fail with UNKNOWN(0x04).
  • Recovery for an unreadable CHDSharp hunk map (#69920). If both compressed CHDSharp attempts produce output whose map cannot be read, the converter now retries once with an uncompressed encode through the separate raw-map writer instead of failing the file. The log warns that this last-resort output is much larger.

Dependencies

  • CHDSharp 1.7.1 brings the two encoder fixes above into the in-process library. SharpCompress 0.50.10, Meziantou.Analyzer 3.0.341 and coverlet.collector 10.1.1 are also current. The code-scanning Path.Join cleanups from the previous maintenance commits are included.

Verification

  • Test suite: 1,314 tests pass (1,285 unit + 29 integration-tagged).
  • JetBrains InspectCode: zero warnings and zero errors.

3.14.0 (2026-11-13)

.NET 11 and current dependencies

  • The app and tests now target .NET 11. CHDStudio builds for net11.0 and net11.0-windows, the test suite for net11.0-windows, and the five libraries add net11.0 to their existing net8.0/net9.0/net10.0 targets. Users install the .NET 11 Desktop Runtime on Windows and the .NET 11 runtime on Linux/macOS; the app remains framework-dependent and single-file.
  • Dependencies refreshed. Avalonia 12.1.6, SharpCompress 0.50.9, Meziantou.Analyzer 3.0.339 and Microsoft.NET.Test.Sdk 18.12.0. C# 15 is enabled through LangVersion. The CI and release workflows, documentation and package tables were updated to match.
  • Bundled tools stay outside the single-file executable. The tool content files carry ExcludeFromSingleFile so the .NET 11 publish keeps chdman/7za/7zz next to the app instead of dropping them from the release bundle.

Platform-native release bundles

  • Linux releases are now .tar.gz. release_<version>_linux-x64.tar.gz and linux-arm64 contain a single CHDStudio/ folder; the archive stores real Unix modes, so CHDStudio and 7zz are executable immediately after extraction.
  • macOS releases are now .dmg. release_<version>_osx-x64.dmg and osx-arm64 mount a CHD Studio <version> volume containing the CHDStudio/ folder, ready to drag to Applications or another permanent folder. The DMG is created with hdiutil on macOS.
  • Windows releases remain .zip. release_<version>_win-x64.zip and win-arm64 keep the flat layout. The release workflow now packages each RID on its native runner and verifies the contents (and execute bits) of all three formats.

Fixes from retrieved bug reports

  • Linux tool permission errors (#69847). A bundled 7zz whose extractor dropped the execute bit is repaired at startup (chmod +x), a tool that still cannot run is skipped in favour of a PATH candidate, and an unstartable tool now names its path and the remedy instead of surfacing a bare Win32Exception.
  • Extraction folder access denied (#69841, #69842, #69843). An output folder that cannot be created is now reported as a folder/permission problem with the folder and the remedy, instead of appending CHD header fields and suggesting the CHD is corrupt. The condition is excluded from bug reports.
  • MDSSharp load failure no longer aborts the batch (#69844). The source-size pre-flight treats an assembly-load failure as non-fatal, counts the input itself and lets the conversion path report a genuine MDS problem.

Verification

  • Test suite: 1,304 tests pass (1,275 unit + 29 integration-tagged).
  • JetBrains InspectCode: zero warnings and zero errors.

3.13.5 (2026-11-08)

Non-standard WAV audio tracks

  • CUE/WAVE tracks that are not 44.1 kHz stereo 16-bit PCM now convert. Both encoders reject other WAVE formats: chdman reports an unsupported sample rate and the built-in CHDSharp encoder says Unsupported samplerate - only 44100 is supported. The app now detects such tracks before encoding and writes a 44.1 kHz stereo 16-bit PCM copy in a private work directory. This covers 22.05 kHz, mono, 8/24-bit and float rips, including Sega CD sets and other non-CD-standard sources. Windows uses NAudio; Linux/macOS use ffmpeg from PATH.
  • Only the affected track is reworked. The cue rework fast path stages the work directory on the cue’s volume and references the disc image in place, so a multi-hundred-MB image is not copied just to normalize one audio track. When a same-volume directory is unavailable or the names require the full copy path, the full work directory normalizes the track there. The user’s original files are never modified.
  • A failed normalization is treated as user data. A corrupt or unsupported WAVE track is logged with the decoder’s reason and no longer creates a bug report.

Code quality

  • Remaining CodeQL code-scanning alerts resolved; code scanning reports zero open alerts.
  • Meziantou.Analyzer updated to 3.0.322.

Verification

  • Test suite: 1,299 tests pass (1,270 unit + 29 integration-tagged).
  • JetBrains InspectCode: zero warnings and zero errors.

3.13.4 (2026-11-08)

Conversion reporting and failed-file retry (issue #15)

  • Conversion batches report real disk usage. The batch measures each successful input’s source set before converting it - the descriptor plus every data file it references, all split-image volumes, or the archive/PBP/CSO file itself - and sums the CHDs that were actually produced. The activity log and completion dialog now show a line such as Space: 22.8 GB of converted source -> 20.0 GB of CHD (saved 2.8 GB / 12.1%). Measuring before conversion keeps the figure correct when Delete originals removes sources during the batch.
  • A CHD that does not shrink is explained. When the output is the same size or larger, the summary says why: the source is already compressed (.pbp, .cso, .isz, .ecm, archives), the image payload is largely incompressible, or the Uncompressed (none) preset is selected. When the output is smaller but Delete originals is off, it also states that the source files still occupy their original space.
  • Failed files are marked and easy to retry. The conversion list has a new Status column (Pending, Converting, OK, Failed, Skipped) and a Select Failed button that selects only the failed rows for the next run. Failed inputs are also written to failed-conversions.txt in the output folder with the source/output folders and one path per file, so the list survives after the activity log is cleared. A clean batch removes the previous report, and a cancelled batch no longer leaves a row stuck on Converting.
  • Version bump: application 3.13.4.

Verification

  • Test suite: 1,296 tests pass (1,267 unit + 29 integration-tagged).
  • JetBrains InspectCode: zero warnings and zero errors.

3.13.3 (2026-11-07)

Retrieved bug reports and input handling

  • 37 retrieved reports fixed or correctly classified. The 20 BatchConvertToCHD and 17 CHDStudio reports in the Bug Report API were reviewed, the application defects behind them were fixed, and user-data/environment failures were classified so they no longer create new reports. The full report log is kept in bugs.md.
  • Alcohol data files resolve from the descriptor’s own directory. MDSSharp now honours a relative directory recorded in the file footer (images\Game.mdf), matches a uniquely decorated .mdf in a subdirectory (Game (USA).mdf), and normalizes Windows/Unix path separators on every platform. A missing file now names the expected file and folder instead of only saying the .mdf was not found. Published as MDSSharp 1.2.3.
  • Malformed cue TRACK types are repaired. A cue that writes TRACK 01 UNKNOWN (or a lowercase accepted type) is normalized from the referenced image’s sector layout — raw 2352 sync/mode, 2048, 2336 or 2324 — before either encoder sees it. Cue rewrites that only change a type no longer copy the whole image set into a temporary directory. If the layout cannot be inferred, the log names the accepted TRACK types instead of filing an application bug.
  • Truncated ISOs are identified precisely. Sector-size validation now reports how many bytes are past or short of a whole sector (the retrieved reports were consistently 4 bytes past 2048-byte sectors) and recommends re-downloading or zero-padding the final sector. The message is classified as user data, so it no longer creates a bug report.
  • Extraction survives header-read failures. A file whose CHD header cannot be read during the parent/child pre-flight is still handed to the extractor, and a file that disappeared after the folder was scanned is skipped with a “rescan the folder” warning. Transient antivirus/indexer locks no longer fail the whole batch.
  • Hash mismatches are explained. Verifying a CHD whose decompressed data does not match the SHA-1/MD5 stored in its header now states that the file is corrupt or was modified and must be re-created from a known-good source; the raw library line is no longer filed as an application defect.
  • Aligned raw inputs are validated for CD conversions too. The early sector-layout warning now covers createcd as well as createdvd, and the post-failure check accepts the 2368/2448-byte subchannel layouts used by Alcohol and CloneCD rips.
  • Version bump: application 3.13.3; library MDSSharp 1.2.3.

Verification

  • Test suite: 1,284 tests pass (1,255 unit + 29 integration-tagged).
  • JetBrains InspectCode: zero warnings and zero errors.

3.13.2 (2026-11-05)

Reliability fixes

  • Fresh temp directories are no longer deleted at startup. CleanupLeftoverTempDirectories now removes only CHDStudio_Temp_* folders older than one hour. A conversion that starts while the delayed cleanup is still pending can no longer have its work/output directory unlinked on Linux, which previously made CHDSharp finish successfully with no visible output file. The missing and zero-length output checks now report the two cases separately and name the path.
  • Uncompressed output is supported for CDs. The Convert and Recompress tabs now honour the Uncompressed preset for createcd; chdman receives -c none and the built-in encoder writes an uncompressed CHD. Laserdisc output still uses AVHuff because no uncompressed A/V codec exists, and that expected fallback is no longer sent to the bug-report API. The chdman codec argument for uncompressed raw/DVD/HDD output was also corrected from NUL characters to none.
  • Cancelling a folder scan no longer raises an unhandled TaskCanceledException. All four file-list scans (Convert, Verify, Extract, Recompress) treat cancellation as a normal outcome, and the Recompress browse handler logs real load failures instead of letting them escape an async void event handler.
  • Version bump: application 3.13.2.

Verification

  • Test suite: 1,260 tests pass (1,231 unit + 29 integration-tagged).
  • JetBrains InspectCode: zero warnings and zero errors.

3.13.1 (2026-11-04)

Verification reliability

  • Verify no longer hangs on large CHD folders. Batch preparation ran on the UI thread and read every CHD’s full header, including its compressed hunk map, before verification started. On a folder with 1,366 CHDs that blocked the window for about 84 seconds, so Windows reported the app as hung and closed it. Preparation now reads only the fixed 76/80/120/108/124-byte header, runs off the UI thread, caches sibling headers for the whole batch, and reuses the resolved parent data in the Verify, Extract and Recompress loops. Cancelling during preparation now works as well.
  • Version bump: application 3.13.1.

3.13.0 (2026-11-04)

CloneCD and trurip descriptors

  • TOC-only .ccd images convert again. trurip and some other CloneCD-derived tools write the whole disc layout as [Entry N] TOC rows and omit the [TRACK N] sections. CCDSharp now rebuilds the track list from those rows (PLBA is the track start, CONTROL bit 2 marks a data track) and reads each data track’s Mode 1/Mode 2 from its first raw sector header. The generated cue is valid and produces a CHD byte-identical to the one made from the set’s own .cue.
  • The zero-track hang is gone. Before this, such a descriptor produced a cue with only a FILE line; chdman 0.289 reads zero tracks and then loops forever (Compressing, nan% complete...) instead of failing, so the batch had to be cancelled. A cue/toc with no TRACK entries is now rejected up front with a clear message, and CCDSharp refuses to write a cue with no tracks or a track without INDEX 01.
  • Cue + CCD sets convert once. When a folder or archive contains Game.cue and Game.ccd for the same disc, the cue is now the input that gets converted and the .ccd is skipped with a log line. Both used to target the same Game.chd, so the second conversion was wasted and was then reported as a duplicate failure.

Libraries and packaging

  • CCDSharp 1.0.3 is published to NuGet with the TOC-only descriptor support and the unusable-cue rejection.
  • Version bump: application 3.13.0; library CCDSharp 1.0.3.

3.12.0 (2026-11-03)

Parent/child CHD support

  • Differential child CHDs are first-class inputs. Verify, Extract, Recompress and the checksum export accept a child plus its parent. Select the parent explicitly or leave auto-detection enabled to match the child header’s parent SHA-1/MD5 against sibling .chd files. A missing or mismatched parent is reported before work starts, and the chdman fallback receives the same parent (-ip for extraction, -op for delta creation) as the built-in reader.
  • A child can be consolidated into a standalone CHD. The new Recompress tab resolves the source through its parent and writes a complete independent image with the selected codec set, so the result works in tools that do not understand parent/child chains.
  • Delta output is verified against its parent. The built-in encoder’s staged delta is checked with Chd.CheckFileWithParent instead of the standalone header check, so a valid child is not mistaken for a parentless/corrupt file and the compatible-codec retry has a real verification failure to act on.
  • Header repair and hashing are parent-aware. Standalone V3-V5 images can have mismatched header hash fields repaired (chdman verify --fix parity), while child images are verified against their parent and never rewritten. Checksum reports use the parent for their whole-image and per-track hashes.
  • Explorer guidance and platform detection. The Explorer detects the disc platform and preselects the matching file-system parser, shows raw and combined hashes, unit counts, header/map/metadata offsets, flags, track pregaps/padding and stored bytes per codec, and rejects child images with guidance to consolidate them first.

Recompression and codec presets

  • New Recompress CHD tab. Batch-recompress existing CHDs, change the codec set, deep-verify the output before replacing the destination, optionally delete the original, and optionally upgrade legacy CD/GD metadata tags. Deep verification is forced when “delete original” is enabled, and children are processed before their parents so a batch containing both cannot lose a parent first.
  • Compression presets for new conversions and recompression. Choose Default (chdman-compatible), Smallest (zstd/cdzs), Most compatible (zlib/cdzl), Auto (per-platform detection) or Uncompressed (raw/DVD/HDD only). Unsupported uncompressed CD/laserdisc output falls back to the compatible set instead of failing opaquely (the Convert tab also logs a warning). The Smallest preset now names four codecs so chdman (four-slot limit) and the built-in encoder produce the same output.
  • Delta CHD creation. New conversions can be written as a child against a selected parent, using the same ParentPath/-op option in both the built-in encoder and chdman.

Conversion and extraction reliability

  • Failed opens no longer leak the PBP file handle. PBPSharp 1.1.8 disposes the file stream when a PBP is rejected, so the file can be moved, deleted or retried immediately; OutOfMemoryException also propagates instead of being reported as damaged user data.
  • Damaged PSAR blocks and PSP EBOOTs are user data, not bug reports. A block that ends before its 16 sectors or an unreadable PBP header is skipped with re-download guidance.
  • Output and temp folder failures are classified. Removed, unready, full, read-only and access-denied drives now produce actionable messages instead of raw I/O exceptions or bug reports, and CHD write errors are treated as output-volume conditions.
  • CHD extraction checks free space first. The target volume is checked against the decompressed size before writing, so a full disk is reported before a partial extraction starts.
  • CHDSharp output is retried once with zlib. If an encoder codec set produces a CHD that fails its own verification, the same source is retried with the most compatible codec set; a valid retry is accepted and the failure is not filed as an app defect.
  • Archive extraction stays responsive. Drive probing and ZIP central-directory sizing run off the UI thread, network drives are filtered before readiness checks, and ZIP extraction reports progress at 10% steps with cancellation checks inside each entry.

Verification

  • Test suite: 1,231 tests pass (1,202 unit + 29 integration-tagged).
  • JetBrains InspectCode: zero warnings and zero errors (suggestions/notes only).
  • PBPSharp 1.1.8 is published to NuGet with the stream-disposal fix.
  • Dependencies: CHDSharp 1.6.3, Avalonia 12.1.5, SharpCompress 0.50.8 and Meziantou.Analyzer 3.0.320 are the current package versions.

3.11.1 (2026-10-31)

PBP extraction - PBPSharp 1.1.7

  • Opening a PBP no longer decompresses disc data. PbpDiscInfo used to decode ISO block 1 in its constructor just to derive IsoSize; a container whose structure was perfectly readable but whose first block tripped the decoder therefore failed the whole file with a bare CorruptFile. IsoSize is now resolved on first use and reports 0 (“unknown”) when it cannot be derived, so the disc opens and extraction reports the real block failure instead of refusing the file.
  • A fully populated ISO index is accepted. The parser rejected the 32,256th index entry as an overflow even though the index area holds exactly that many, so a disc that filled it could never be opened.
  • PBP failures now carry a cause. Every invalid block-length rejection records the block index, block count, disc, disc id and stored length in PbpDiagnostics; PbpFile.Open records the exception type and message instead of returning only an error code. The Failed to open PBP file and Failed to extract PBP file messages include that detail.
  • Fewer false bug reports. A damaged .ecm download (whole-image checksum mismatch) and a laserdisc AVI whose video/audio codec is not YUY2/VYUY/UYVY now log without filing a bug report; chdman’s avhuff has the same codec restriction, and the messages already tell the user what to do.

CHDSharp diagnostics

  • CHDSharp’s internal log now reaches the application log. Chd.LoggerFactory was never assigned, so the library logged into a null logger and its hunk-level diagnostics (failing hunk, codec, compressed length, file offset and the codec’s own reason) were discarded. The new ChdSharpLoggerFactory bridges them to Serilog and attaches the captured detail to the messages that previously said only Decompression failed / Invalid or corrupt data.
  • CHDSharp warnings no longer file bug reports on their own. The bridge writes the library’s Warning level at Serilog Information, because CHDSharp also uses Warning for user-data notices such as “track has subcode data; it will be omitted”. Genuine failures are still reported by the app’s own classified error message, now with the hunk detail attached.
  • Stale diagnostics cannot leak between files. Each CHD open and staged-output validation clears the previous captured warning first; a ChdFile.Open failure now includes the header/map reason too.

Security and dependencies

  • CodeQL code-scanning findings fixed. Archive entry paths are validated inline before extraction (the Zip Slip guard is now visible to CodeQL), the single-instance mutex, CHD container, CSO stream and chdman speed counter are disposed on every failure path, and empty catch/loop blocks were replaced with documented handling.
  • CHDSharp 1.6.0 is the bundled encoder/reader.
  • CodeQL ignores generated bin, obj, publish, dist and TestResults output.

Verification

  • New regression tests cover the lazy ISO size, a fully populated index area, corrupt block diagnostics, the CHDSharp log bridge and the new bug-report exclusions. The full suite passes (1182 tests: 1153 unit + 29 integration, 59 test classes); CI runs the 1153 unit tests.
  • JetBrains InspectCode reports zero warnings and zero errors.

3.11.0 (2026-10-29)

PBP extraction - PBPSharp 1.1.6

  • Blocks that carry trailing bytes now extract. A PSAR block is accepted as soon as it has produced one full 16-sector block; inflation no longer parses whatever follows it. Writer padding, the next stream’s first bytes when an index length overstates a block, and a stream that never sets BFINAL no longer turn a complete block into a DecompressionError (“broken uncompressed block”, “Unknown block type 6/7”, “Encountered invalid codelength 0”, “Illegal rep dist code”, …). This fixes the batch of PBP failures reported on 3.10.1 and matches the popstation/PSX2PSP/iPoPS reference readers, which stop when the output is full.
  • A block whose index length under-states its stream is read up to the next entry. When the following index entry starts further away, that distance is the block’s real extent, so a length recorded before the writer’s final flush no longer fails with a short stream.
  • A zero-length compressed index entry is rejected. It is a corrupt index, not an under-stated stream; reading the following block’s data as this block would silently shift the extracted image.
  • Block diagnostics now include the number of bytes actually read (read= next to length=) and a short block reports how many of the expected bytes it produced, making a genuinely truncated block distinguishable from a padded one.

Encoder failure reporting

  • A C++ runtime abort inside chdman is recognised from its output. libc++abi: Pure virtual function called! and the terminate called ... family exit with a positive code on Windows, so the crash was previously treated as a normal conversion error and re-filed for every following file. It now switches the rest of the batch to the built-in encoder and the raw runtime line is no longer sent as a bug report.
  • A failed CHDSharp fallback keeps its real cause. The generic “the built-in CHDSharp encoder did not produce a usable output” message is only logged when the encoder failed without logging a reason; a truncated image, a cue both encoders reject, or an I/O error now appears as the specific, classified message instead of being hidden (and reported twice).
  • CHDSharp output is checked before it replaces the destination. A missing, empty or unreadable staging file is reported as a failed conversion instead of being moved into place. Any validation failure, not only an I/O error, is treated as a failed conversion.
  • Timeouts and duplicate outputs keep their own message. They no longer add a second, generic “did not produce a usable output” line after the specific timeout or duplicate-output notice.

CI, security and release engineering

  • CodeQL code scanning runs for C# and the GitHub Actions workflows with the security-and-quality query suite on every push/PR to master, weekly, and on demand.
  • Dependency Review blocks pull requests that introduce high or critical vulnerable dependencies.
  • OpenSSF Scorecard runs weekly and on branch-protection changes; results are published and uploaded to code scanning.
  • Dependabot keeps NuGet packages and SHA-pinned GitHub Actions up to date with weekly grouped pull requests.
  • GitHub Code Quality is enabled for C# and the Node helper scripts.
  • Repository hardening: secret scanning and push protection, private vulnerability reporting, SHA-pinned Actions, approval for external fork workflows, a read-only default GITHUB_TOKEN, and a Protect master ruleset that requires pull requests plus the CI/CodeQL/Docs/Dependency Review checks (admins can bypass it for direct maintenance).
  • Release provenance: every release zip gets a signed build-provenance attestation; verify it with gh attestation verify release_<version>_<rid>.zip --repo purelogiccode/CHDStudio.
  • Release verification: the release workflow now checks that all six RID zips are attached before it finishes.
  • Documentation: a new security and quality page, side menus for Pages and the wiki, badges on every README, and updated 3.11.0 package tables.

Verification

  • New regression tests cover padded unfinished blocks, an index length that lags the next entry, a zero-length compressed entry, C++ runtime abort detection and the new bug-report exclusions. The full suite passes (1175 tests, integration included); CI runs the 1145 unit tests.
  • JetBrains InspectCode reports zero warnings and zero errors.

3.10.1 (2026-10-21)

Multisession cues now encode byte-identically to chdman

  • CHDSharp 1.5.2 fixes two cue-layout divergences from chdman. A single-bin multisession cue whose REM LEAD-OUT sits at or before the session’s last track (REM LEAD-OUT 00:00:00 is common in real IsoBuster/ImgBurn rips) used to trim that track to a negative length and abort the conversion with the session lead-out values leave a negative length (-11400 frames). MAME compares a uint32_t frame count against a signed int32_t end frame, so the value wraps and the trim is skipped; the built-in encoder now performs the same unsigned comparison and reproduces chdman’s layout byte for byte.
  • A pad larger than its track no longer zeroes the track. The CD reader now mirrors chdman’s unsigned pad offset, so the track is read from the source file instead of being written as zeros, and the GD-ROM data/pad/split ordering matches chdman as well.
  • The cue work directory no longer repairs these cues. The 3.10.0 workaround that dropped an unusable REM LEAD-OUT line is gone: the cue reaches the encoder unchanged, so chdman and the built-in encoder stay identical.

Cue reference resolution

  • Two FILE lines that name different files can no longer collapse onto one image. A missing track whose name matched another track’s file was silently resolved to it, turning a split-track cue into a single-bin one and making two tracks start at the same position - which both encoders reject with Unable to determine size of track N, missing INDEX 01 markers?. The second reference is now reported as missing, with a message that names it.
  • That failure is classified as user data and excluded from automatic bug reports, with guidance telling the user to re-create the cue.

Verification

  • New regression tests cover the zero lead-out, a lead-out before the track, the missing-file collapse and the .-prefixed repeated FILE line, plus an end-to-end encode of a zero-lead-out cue through the built-in encoder.
  • Test suite: 1162 tests (1133 unit + 29 integration) across 58 test classes. CHDSharp 1.5.2 ships 454 encoder + 4214 library tests and 905 battle-harness checks, all green.

3.10.0 (2026-10-21)

Large and multi-disc PBP files

  • PBP disc offsets are 64-bit throughout the parser. PSAR and per-disc positions were widened before every seek, so a multi-gigabyte PBP whose later disc lies beyond the signed 32-bit boundary no longer fails with a bogus IoError.
  • Truncated multi-disc PBPs are identified before extraction. A disc position that points past the end of the file is reported as TruncatedPsar instead of seeking to a negative offset or producing a partial BIN.
  • PSAR blocks get a second inflater. Blocks rejected by SharpZipLib’s stricter state machine are retried with the .NET zlib implementation; a block is accepted only when it produces exactly one complete 16-sector block.
  • PBP extraction checks free space first. The estimate uses each disc’s declared ISO size plus a margin, and a nearly full temp drive gets an actionable message before a multi-gigabyte extraction starts.

Damaged CHDs and clearer diagnostics

  • A CD CHD with missing or damaged track metadata is rejected with a precise message instead of a bare Chderrinvaliddata. The diagnostic log records the track count and metadata tags so a damaged CHT2/CHTR section is distinguishable from a non-CD image.
  • The message and log guidance tell the user to re-create the CHD from its cue/bin. This is classified as damaged user data, not an application bug.
  • Malformed cue lead-out and truncated-image errors now include recovery guidance, and both are excluded from automatic bug reports.

Startup and bug-reporting reliability

  • A missing native rendering library no longer produces an unhandled startup crash. The app exits with a clear message naming libHarfBuzzSharp, libSkiaSharp and the platform native backend, and writes startup-error.log when the console is unavailable. This normally means the release zip was not extracted in full; copying every file beside the executable fixes it.
  • More routine conditions are excluded from the bug tracker: missing native libraries, temp-directory cleanup held open by antivirus/indexers, damaged CHD track metadata, malformed cue lead-outs and truncated source images.
  • InspectCode warnings are cleared. The solution now reports zero warning-level inspections; a warning-level InspectCode.log is available for local verification.

Release engineering, documentation and verification

  • All six release bundles are produced from the same version: win-x64, win-arm64, linux-x64, linux-arm64, osx-x64 and osx-arm64.
  • Each zip contains the framework-dependent single-file app, its native libraries, the correct architecture’s tools, LICENSE.txt, ReadMe.md and WhatsNew.md. No .NET runtime is embedded; users install the .NET 10 runtime.
  • Linux/macOS zips preserve Unix execute bits (0755 for CHDStudio and 7zz), and the release workflow verifies them before publishing.
  • Pages and wiki documentation are deployed from the same docs/ tree, with a dark Just-the-Docs side menu and a wiki _Sidebar.md. Pull requests build the Pages site; master deploys it and syncs the wiki.
  • The documentation index, ReadMe badges and package tables now carry the 3.10.0 release data.

Package updates

  • CHDSharp 1.4.5 -> 1.5.0
  • NAudio 3.2.0 -> 3.3.0
  • SharpCompress 0.50.5 -> 0.50.7
  • Meziantou.Analyzer 3.0.303 -> 3.0.315

Verification

  • Build: 0 warnings, 0 errors.
  • Tests: 1157 passed (1128 unit + 29 integration).
  • jb inspectcode --severity=WARNING: 0 findings.

3.9.5 (2026-10-16)

Encoder library: CHDSharp 1.4.4 / 1.4.5

  • Redump Dreamcast multi-cues convert byte-identically to chdman. A .cue holding both images of a GD-ROM (REM SINGLE-DENSITY AREA / REM HIGH-DENSITY AREA) previously produced a plain-CD layout; the parser now applies MAME’s is_gdicue rules — pregaps move onto the previous track as split regions, the last single-density track pads up to LBA 45000, and the CHD is stamped with CHGD GD-ROM metadata. Verified against a real 1.19 GB Redump multi-cue: identical SHA-1 and metadata to chdman createcd, and chdman extractcd on the app’s output restores every track byte-for-byte.
  • Multisession cues (IsoBuster/ImgBurn/Redump/DiscImageCreator) are laid out like chdman. REM SESSION / REM LEAD-IN / REM LEAD-OUT and session REM PREGAP are now honored: single-bin images trim the session’s last track at the declared lead-out and turn the inter-session dummy data into zero-filled pad frames (with DiscImageCreator’s .img no-padding heuristic); multi-bin images get the synthesized standard lead-in/lead-out values. An inconsistent lead-out that would go negative is rejected with a clear error instead of being encoded as a silently wrong image. Byte-parity is pinned by chdman createcd fixtures for all three layouts.
  • cdrdao .toc files with a bare START line convert. MAME accepts a numberless START (no pregap); the built-in parser rejected it with Malformed START command, failing every .toc chdman converts. Both START forms now produce byte-identical output to chdman.

cdrdao .toc staging and malformed-cue fixes (#68261, #68249)

  • A staged .toc keeps its extension. When a descriptor needed the work directory (UTF-8 BOM, non-ASCII names, …) it was always renamed to game.cue; cdrdao syntax (TRACK MODE2_RAW, DATAFILE …) then reached the CUE parser in both encoders and failed with Malformed TRACK command: TRACK MODE2_RAW. The work descriptor is now game.toc for .toc input, and its DATAFILE/AUDIOFILE references are copied and rewritten with their SWAP/offset/length arguments preserved.
  • Backslash-escaped quotes and missing BINARY tokens are repaired. A cue written as FILE \"track01.bin\" (escape backslashes, no type token) could not resolve its bin and made the fallback report Malformed FILE command. The normalizer now unescapes the name and infers the missing type from the extension, producing a canonical cue that both encoders accept.
  • A descriptor that describes more data than its file holds (truncated rip) is documented and classified. chdman reports Error reading input file, the built-in encoder Unexpected end of file — both reject the same files, so it is a user-data condition; the troubleshooting page explains it and it files no bug report.

chdman crash handling (#68281)

  • One conversion-time crash switches the batch to the built-in encoder. The startup probe only runs chdman help; a chdman that dies later with STATUS_ILLEGAL_INSTRUCTION (a CPU without the build’s SIMD extensions) used to crash again for every file. The first negative exit code now sets a per-batch flag: the remaining files are encoded in-process, and the crash summary is excluded from bug reports as an installation problem (matching the startup-check policy).

Fewer false bug reports (#68256, #68260)

  • chdman exited with code N and its output is not a valid CHD (…); discarding it. — the intermediate validation notice for a partial output no longer files a report; the classified final message still reports the real cause.
  • Source file appears to be corrupt: … (a CRC/data error while copying a source for the temp retry) is a user-data condition; it is now excluded from reports and logs the underlying exception to the local log file.

Housekeeping

  • Package updates: CHDSharp 1.4.3 → 1.4.5 (two parity releases, see above), NAudio 3.1.0 → 3.2.0.
  • Test suite: 1146 tests (1117 unit + 29 integration) across 58 test classes; CHDSharp 1.4.5 ships 451 green tests, including the chdman byte-parity suites for the three descriptor fixes.

3.9.4 (2026-10-13)

CD extraction writes cooked sectors again (#11)

  • Extracted BINs are Redump-identical again. The built-in reader called CHDSharp with cooked:false, writing full 2448-byte raw frames (data plus 96-byte subcode) instead of chdman’s cooked DataSize bytes per frame. A single-track MODE2_RAW disc extracted as 546,178,176 bytes (SHA1 072ea00f...) instead of 524,759,424 bytes (SHA1 9b4c1b51...) while the CUE looked identical. Extraction now passes cooked:true, matching chdman extractcd byte-for-byte (subcode omitted, audio byte-swapped). Reproduced against Armored Core (USA) (Rerelease).
  • Verify passing for the bad rebuild is by design. A CHD built from the oversized BIN is internally consistent (232,218 frames instead of 223,112; 267,908,519 bytes), so both chdman verify and CHDStudio Verify report success. Verify checks the header hashes, not Redump: the header SHA1 (6dc714d9...) and the file SHA1 (3e1b20c5...) differ from 7-Zip’s file hash by design. Delete BINs/CHDs produced by the broken extraction and re-extract.

Housekeeping

  • Package updates: Meziantou.Analyzer 3.0.299 → 3.0.303, Microsoft.NET.Test.Sdk 18.10.1 → 18.11.0, VideoGameFileSystemParser 1.3.0 → 1.4.0.
  • Test suite: new ExtractCd_CookedMatchesOriginalBin_RawAddsSubcode regression test; 1105 unit tests pass (Category!=Integration).

3.9.3 (2026-10-11)

Cue WAVE tracks with an over-declared header now convert (#68232–#68237)

  • A WAV whose header claims more audio than the file contains no longer fails the conversion. Both encoders trust the data chunk’s size field, so a file that is short of its declared length — the widely distributed Policenauts Saturn English patch ships exactly this shape, one CD frame (2,352 bytes) over-declared on all three discs — made chdman abort with Error reading input file (...wav) and the built-in CHDSharp fallback report Unexpected end of file (...wav). The app now detects the over-declaration before encoding and clamps the data and RIFF sizes to the bytes actually present, so the conversion reads the audio that exists. The repair happens on a private copy; the user’s files are never modified.
  • Only the damaged WAV is copied. The repaired work set is staged on the cue’s own volume and references every other file (including multi-hundred-MB disc images) in place by relative path. When a same-volume location is unavailable it falls back to the full copy-based work directory, which repairs the WAVs there.
  • Both encoders produce the same result for a repaired set: a byte-identical CHD from chdman and the built-in CHDSharp encoder (verified against the real Policenauts disc 1 files).

Housekeeping

  • PbpFileIntegrationTests now early-returns when its local samples are absent, matching every other integration group (§11.5); without samples the tests report as passed instead of failing on Assert.NotEmpty().
  • MDSSharp published as 1.2.1 with repository metadata pointing at the renamed CHD Studio repository; no functional changes.
  • Package updates: Avalonia 12.1.3 → 12.1.4, SharpCompress 0.50.4 → 0.50.5, Meziantou.Analyzer 3.0.294 → 3.0.299, xunit.runner.visualstudio 4.0.0 → 4.0.1.
  • Test suite grew to 1133 tests (1104 unit + 29 integration) across 58 test classes, including WavHeaderSanitizer unit coverage, the CueWorkDirectory repair paths and a real-chdman end-to-end regression test for the repaired cue set.

3.9.2 (2026-10-05)

Duplicate outputs are skipped before conversion

  • A duplicate input is stopped before any encoding starts. When two inputs in one batch resolve to the same output CHD and the collision only becomes visible once an archive is extracted (its product is named after a file inside the archive), the duplicate is now skipped immediately; previously it was converted in full and the finished product discarded. The first product is kept exactly as before — the only difference is that no time is spent encoding something that will not be used.
  • The duplicate notice can no longer become a bug report. "<name>.chd was already produced earlier in this batch; keeping the first one." is now logged at information level instead of warning. Combined with the 3.9.1 exclusion pattern (kept as a safety net), this ends the false reports that 3.9.0 clients sent for every duplicated game in a batch.

Reliability

  • Screenshot capture no longer risks a disposed bitmap. The F8 capture now hands the RenderTargetBitmap to a dedicated SaveScreenshot overload owned by ScreenshotService, instead of capturing it in a delegate that escaped the using block that disposes it.

Housekeeping

  • MDSSharp cleanups: the index00Tracks null check is expressed as index00Tracks?.Contains(track.Number) ?? pregapsInFile, and long locals use var where the type is apparent.
  • PBPSharp and test-project cleanups of the same kind; the throughput test documents its deliberate captured-counter lambda with a SuppressMessage.

3.9.1 (2026-10-05)

PBP extraction — PBPSharp 1.1.4

  • PBPs whose compressed blocks were written without a final deflate block now extract. Some authoring tools flush each PSAR block with Z_SYNC_FLUSH instead of finishing the stream, so the last deflate block never sets BFINAL. SharpZipLib’s stream wrapper reported that as “Unexpected EOF” and the whole file failed with DecompressionError; the raw inflater is now driven directly and accepts a complete 16-sector block, while a block cut short still fails. Verified byte-identical against real Brave Fencer Musashi and Ridge Racer PBPs across compression levels (stored, default and maximum).
  • A block cut short can no longer be accepted as a short block. The decompressed output is bounded to exactly one 16-sector block, so a stream that ends before a full block, or expands past one, is rejected with DecompressionError instead of silently shifting every later block in the extracted image.

Fewer false bug reports

  • The by-design duplicate-product notice is no longer reported. <name>.chd was already produced earlier in this batch; keeping the first one. appears when two inputs resolve to the same CHD (typically an archive whose contents were not known at the collision preflight); the first product is kept, as intended.
  • CHDSharp device and permission errors are classified as environment conditions. A drive that was unplugged, a failing disk, a full destination or denied access now logs CHDSharp could not read or write the image: ... and is excluded from bug reports; only genuine encoder defects still report.

Release bundles for all six platforms

  • The Release workflow now publishes and attaches one zip per runtime identifier: win-x64, win-arm64, linux-x64, linux-arm64, osx-x64 and osx-arm64 — all framework-dependent single-file builds, so users install the .NET 10 Desktop Runtime on Windows and the .NET 10 runtime on Linux/macOS. Each zip carries only its own platform’s tools (7za/chdman on Windows, 7zz plus 7-Zip-License.txt on Linux/macOS) and the platform’s native Avalonia libraries, and the packaging script verifies every required file before zipping.
  • Linux and macOS are no longer source-only. They now get the same release downloads as Windows, matching the already cross-platform application.

Housekeeping

  • Roslynator.Analyzers bumped 5.0.0 → 5.0.1.
  • Test suite grew to 1121 tests (1093 unit + 28 integration), including regression coverage for unfinished deflate streams, truncated PBP blocks and a zero-length stored index entry.
  • PBPSharp published as 1.1.4; the README, library reference, bug-reporting guide and troubleshooting guide document the new block handling and exclusions.

3.9.0 (2026-10-03)

Renamed to CHD Studio

  • The project, application and executable are now CHD Studio (CHDStudio): the GitHub repository moved to https://github.com/purelogiccode/CHDStudio (the old URL redirects), the solution is CHDStudio.sln, the project folder is CHDStudio/ and the Windows executable is CHDStudio.exe. The five NuGet libraries keep their names; their repository links point at the new URL from the next package release.
  • Logs and screenshots now live under %LocalAppData%\CHDStudio (old %LocalAppData%\BatchConvertToCHD logs are not migrated); the temporary work folder fallback is now CHDStudio_Temp.

Avalonia becomes the application front end

  • The Windows-only WPF front end is gone; the cross-platform Avalonia app is now the base project and builds CHDStudio.exe. It runs on Windows, Linux and macOS; official release zips remain the two Windows architectures.
  • The application project was renamed BatchConvertToCHD.Avalonia → CHDStudio, so the folder, project and executable all carry the product name.
  • Shared code and the bundled tools moved into CHDStudio/, so the project no longer links files from another folder. Windows keeps the bundled chdman/7za/CHDSharp; Linux and macOS discover them (plus ffmpeg for MP3 tracks) on PATH.
  • UI parity with the old WPF build was restored: Alt+letter button mnemonics, click-to-sort grid columns with the original header tooltips, and accessibility names on the main controls.
  • F8 screenshots are now window-scoped: while the app window is focused, F8 captures it with RenderTargetBitmap on every platform (previously a system-wide hotkey captured the foreground window on Windows only). Screenshots are saved into the screenshots folder inside %LocalAppData%\CHDStudio, beside the logs folder, with a screenshots folder next to the app as fallback.
  • The main window has a Donate button (left of About) that opens the project’s donation page, and the About window now scrolls so every acknowledgement is reachable on short screens.
  • The Close button always closes the app again: shutting down from inside the window’s Closing event re-entered the close in a loop, so the window never actually closed while an operation was idle.
  • The activity log is batched and capped: log lines are flushed in batches instead of one dispatcher call per line, and the on-screen text is capped, so a very large log can no longer freeze the window. The Explorer tab now hides the log panel entirely instead of leaving a clipped strip of it over the window edge.
  • The About window is complete: it now carries the full third-party acknowledgements (CHDSharp, CHDMAN, Avalonia, the in-house image libraries, VideoGameFileSystemParser, SharpCompress, NAudio, Serilog and 7-Zip), a License section (GPL-3.0 with a link), the copyright line, and donate/release links.

New CHD Explorer tab

  • Browse a CHD like a disc: pick a .chd, choose the file-system parser matching its console/system (35 formats, PlayStation auto-detection by default), and navigate the folder tree in a grid. Double-click a folder to open it, a file to extract and open it, or use Extract to save a folder to disk.
  • Image Info: an expander under the parser drop-down shows a read-only report built from the CHD header and map — version, hunk/unit sizes, logical size, image type (CD/GD-ROM/DVD/HDD/A-V), codecs, metadata tags, the CD/GD-ROM track table and the per-codec hunk distribution (the scan is capped at the first 1,000,000 hunks for very large images).
  • Backed by VideoGameFileSystemParser 1.3.0 (https://www.nuget.org/packages/VideoGameFileSystemParser), rebuilt against CHDSharp 1.4.3 and published for this release.

Laserdisc A/V support

  • .avi laserdisc captures convert with createld: the built-in CHDSharp encoder assembles every video frame into MAME’s raw chav layout and compresses it with the avhu codec (delta-RLE Huffman video + per-channel mono FLAC audio), one frame per hunk, exactly like chdman createld; the Force CD/DVD overrides are ignored for an AVI.
  • A/V CHDs extract back to AVI in-process: the Extract tab gained a Laserdisc (.avi) format, and Auto-detection now reads the CHD’s own flags — a CHD with no CD/DVD/HDD metadata is an A/V image and is extracted with CHDSharp’s MAME-parity ExtractLaserDisc writer (YUY2 video + PCM audio). chdman extractld/extractraw remain the automatic fallback when the library cannot decode the CHD.

Verification checksum reports

  • Optional <name>.checksums.txt next to each verified CHD (checkbox on the Verify tab): one whole-image SHA-1, CRC-32 and XXH3-64 plus per-track SHA-1, CRC-32 and XXH3-64 for CD/GD-ROM images. The whole-image hashes are computed over the decompressed image, so they match a hash of the extracted data rather than the CHD header’s combined hash. The report follows the CHD into the Success folder, and a report that cannot be written is a warning — never a verification failure.

Live tool output, progress and read-ahead

  • Every chdman output line now reaches the activity log, progress included, with completion lines (“Compression complete”/”Extraction complete”/”final ratio”) marked with a check. The on-screen log caps a single line at 2,000 characters and appends at most 200 lines per flush, so a chatty tool cannot stall the UI, and the file sink rolls at 10 MB.
  • The built-in CHDSharp encoder and reader log progress every 10% for conversion (ratio=), verification, extraction and checksum hashing, so long library operations visibly keep running without one line per hunk.
  • Extraction pre-decompresses 16 hunks in the background (ChdFile.ConfigureReadAhead), so sequential reads overlap the disk writes on the single-file (DVD/HDD) and multi-track (CD/GDI) paths.

Built-in CHD encoder — no CHDSharp CLI

  • CHDSharp now runs in-process (the NuGet library, not the bundled CHDSharp.exe CLI, which has been removed). On Linux and macOS it is the encoder, so those builds need no bundled native tool; on Windows the bundled chdman remains the primary encoder and the built-in CHDSharp is the automatic fallback. It mirrors chdman’s commands and defaults (createcd/createdvd/createhd/createraw) and produces byte-identical CHDs.
  • Conversion can no longer fail for a missing encoder; a missing Windows chdman.exe only logs that the built-in encoder will be used. The status bar shows CHDSharp as always available.
  • Official 7-Zip 26.03 console binaries for Linux and macOS are now bundled (7zz, static builds on Linux and a universal binary on macOS) and copied next to the app for the matching runtime; 7-Zip-License.txt is included. Windows keeps 7za.exe.

Reliability fixes

  • CSO v2 / ZSO images decode correctly. The reader applied CSO v1 index semantics to version 2: LZ4 blocks (high bit set) were returned as raw compressed bytes and stored or deflate blocks were pushed through the LZ4 decoder. Blocks are now classified the way the format defines them — compressed only when smaller than a full block, with the high bit selecting LZ4 over deflate — and a short final stored block is zero-padded instead of failing.
  • A truncated CHD is never accepted as a success. When chdman exited nonzero but left a non-empty file (the classic disk-full case), the app treated it as success and could delete the source. The output is now validated with the built-in reader first, and a partial file is discarded.
  • Deleting originals never reaches outside the input folder. A descriptor naming ..\..\other\game.bin (or an absolute path) could make the batch delete files it never converted; referenced files outside the selected input folder are now kept and reported.
  • Archived discs keep their folder structure. Two same-named discs in different archive subfolders (Disc1/game.cue, Disc2/game.cue) both produced game.chd, and the second silently replaced the first; the archive’s internal path is now preserved. The same-batch duplicate guard keeps the first product and reports any remaining collision.
  • Extracting next to an existing disc set no longer replaces it. The “extract into a subfolder instead of overwriting” rule now also covers cue/gdi sets (checking their BIN too) and applies to the chdman fallback, not just the built-in reader. Cancelling a fallback extraction now kills the chdman child instead of leaving it running.
  • A PBP whose volume descriptor declares zero sectors no longer extracts an empty BIN and reports success; every index entry is written when the size is unknown.
  • Expected network and user conditions no longer file bug reports: an update check that times out (now bounded to 30 seconds instead of the shared client’s 100) and a CHD Explorer parse failure caused by choosing the wrong file-system parser are logged as network/user conditions instead of being sent to the bug tracker.
  • PBP I/O failures carry the underlying OS error (PBPSharp 1.1.3): an open or extraction that fails with IoError records the file, the OS message and whether it was a read or a write, so a locked file, a failing disk or a full drive is identifiable from the log and the bug report.
  • CD-R/CD-RW MDS v2 descriptors read their track lengths from the extra block like pressed CDs, and a .ccd claiming a huge TocEntries value can no longer make the parser allocate unbounded tracks.
  • MP3 decoding on Linux/macOS is cancellable: ffmpeg is awaited with the operation token and killed on cancel instead of blocking the window in an uncancellable WaitForExit().
  • Real .mdx containers are read: the MDS v2 parser applied the 1 MB v1 descriptor cap before checking the version, so every genuine single-file MDX (which embeds the whole image) was rejected, and the whole file was loaded into memory. The header is now read first, the descriptor is decrypted/decompressed from a stream, and track data is decoded per footer: the footer’s track_data_length (the sectors actually stored) is used instead of the extra block’s logical length, so a pregap kept in the data file is no longer truncated, and a track split across several data files decodes every footer in order. The decoder also streams groups instead of buffering whole tracks, and a compressed descriptor that legitimately expands beyond the file size is no longer rejected.
  • chdman createcd no longer receives an invalid -us 2352 switch. A cue referencing a .raw track made the Windows primary encoder exit with “Option ‘-us’ not valid for this command” and silently fall back; chdman derives the unit size from the cue’s track types, so the switch is now only passed to createraw — including when a .raw input is combined with Force CD or Force DVD, which previously still appended it to createcd/createdvd.
  • The built-in encoder rejects inputs chdman refuses. A createdvd/createhd/createraw input whose size is not a whole number of units is now rejected up front instead of dropping the partial trailing unit and reporting success (after which the source could be deleted).
  • Explorer extraction never replaces existing files: extracting an entry into a folder that already holds a file or folder of the same name now lands in a numbered subfolder, the same isolation the Extraction tab applies; a staged directory is renamed into place in one step when the destination is free, and a temp extraction opened in a viewer is deleted after the viewer exits instead of 30 seconds later.
  • The same-batch duplicate-output guard now covers the built-in encoder too, and a reservation is released when the final move fails, so a failed product cannot block a later input that resolves to the same CHD.
  • Temp folders no longer litter every drive. GetBestTempDirectory prefers the system temp folder whenever its path is safe for chdman and its volume has room, instead of always picking the roomiest drive and creating CHDStudio_Temp at its root; the drive-root fallback is only used when %TEMP% is unsafe or too small, and an emptied CHDStudio_Temp folder is removed after use and at startup (stale Explorer extractions older than a day are cleaned up as well).
  • The speed card shows chdman’s throughput. The write-speed sampler measured the application’s own process I/O, which stays near zero while the bundled chdman (a separate process) does the writing, so the card read 0.0 MB/s for every chdman conversion and looked stalled. It now samples the chdman child process, while the in-process CHDSharp encoder keeps using the app’s counter.
  • The activity log keeps its mode instructions when the tab changes: the clear is now synchronous, so the welcome/ready lines queued by the same selection change are no longer discarded.
  • Completion dialogs are awaited: the batch summary is shown before a close requested during the run is honoured, and the encoder notice is dismissed before the update prompt appears.
  • Smaller fixes: a file-system-watcher history clear is atomic with event recording; a Windows drive root is no longer trimmed to a drive-relative path when building a temp folder; the ISZ chunk table and PBP SFO string length are bounded before their int casts; a CCD .img with a partial trailing sector is rejected instead of silently dropped; archive bin sets are grouped case-sensitively on case-sensitive file systems; a bare .bin with a companion cue now converts through the cue; 7-Zip arguments are passed via ArgumentList so quotes in paths cannot inject switches; the bug-report throttle’s safety-net timer can no longer clear the flag of a newer send; a corrupt CSO index, MDX descriptor or ISZ chunk table is rejected before a huge allocation.
  • A laserdisc AVI that fails both encoders is no longer misdiagnosed: the post-failure sector-alignment diagnostic exempts .avi, so the real encoder error is shown instead of “file size is not divisible by any standard sector size”.
  • A temp extraction opened in an external viewer is no longer deleted too early on Linux/macOS: xdg-open/open exit as soon as they hand the file to the viewer, so a quick exit now gets the fixed grace delay instead of removing the file while the viewer is still starting.
  • A CSO whose header declares zero uncompressed bytes no longer “extracts” to an empty ISO and reports success; it is reported as an invalid header. A temp directory is only accepted as writable when it really is, and a descriptor’s referenced-file deletion check compares paths case-sensitively on case-sensitive file systems.
  • MDS v2 footers are read like the format reference: a track whose footer count is zero but that carries a footer is decoded from that footer instead of being treated as uncompressed, and a later fragment of a split track that names no data file is reported instead of silently reading the first data file again.
  • MDS v2 pregaps are resolved per track: the footer’s stored length tells whether each track’s pregap is in the data file (length = missing, length + pregap = stored, the layout libMirage validates). An image that keeps one track’s pregap and omits another’s is now rebuilt selectively — only the missing pregaps are materialized as zeros — so every INDEX 00/INDEX 01 lands at the LBA the descriptor names instead of the cue drifting after the first omitted pregap. The first track’s pregap before LBA 0 is never materialized, matching libMirage’s NULL pregap for track 1.

Housekeeping

  • Version bumps: application 3.9.0; libraries CCDSharp 1.0.1, CSOSharp 1.0.1, ISZSharp 1.0.2, MDSSharp 1.2.0 (MDS v2/MDX), PBPSharp 1.1.3; Meziantou.Analyzer 3.0.294.
  • Release script strips native .pdb debug symbols from the zip and verifies the Avalonia native libraries are present.
  • Docs, AGENTS.md and CI updated for the new base project; tests now reference the Avalonia assembly.
  • Test suite grew to 1116 tests (1088 unit + 28 integration), including new CSO v2 stored/LZ4/deflate, PBP zero-size, corrupt-CCD, MDX container, encoder divisibility, IoThroughputCounter, checksum-report, Image Info, CHDSharp-progress, MDS v2 per-track pregap and createld regression tests (with a committed laserdisc AVI fixture).
  • The docs/ folder is the single source of truth for both documentation homes: GitHub Pages (Jekyll/just-the-docs navigation) and the GitHub wiki (docs/_Sidebar.md side menu, index.md → Home.md, WhatsNew.md synced from the repository root).

3.8.0 (2026-09-19)

Multi-part RAR archives now convert end to end (#67305)

  • A .partNN.rar set is decoded from its first volume, whether the batch lists the first part or a later one. SharpCompress can only follow the whole set when it is opened by path through the first volume, so a later part is redirected to the first volume found beside it; when the first volume is missing the file is skipped with a message naming it instead of failing inside the decoder. This supersedes the 3.7.0 note that multi-part RAR still needed manual extraction.
  • RAR sets renamed to .001 are extracted by content instead of being refused, because the extension hides what the content says.
  • A set is offered once: only the first volume stays in the batch, filtered both at the folder scan and again before the batch starts, so a multi-part RAR is no longer extracted once per volume.
  • Malformed archives are classified as data, not app bugs: SharpCompress decoder crashes (NullReferenceException, ArgumentOutOfRangeException, IndexOutOfRangeException) on corrupt data no longer trigger the temp-copy retry or an automatic bug report.
  • Volume discovery covers new-style .partNN.rar, numbered .001 sets and old-style .rar + .r00 volumes, and the disk-space preflight measures the whole set. A real WinRAR-built multi-volume fixture is committed so the path is tested without WinRAR at test time.

Alcohol 120% support upgraded — MDSSharp 1.1.0

  • The library was renamed Alcohol120Sharp → MDSSharp and published as MDSSharp 1.1.0 (https://www.nuget.org/packages/MDSSharp) with package metadata, a README and an icon matching the other embedded libraries.
  • Pregaps are now described: when the descriptor records pregap and track length, the generated cue carries INDEX 00 if the data file already contains the pregap sectors, and pregaps the file does not contain are rebuilt as zeros so the cue can still express them without shifting the tracks that follow.
  • Multi-file descriptors are joined in track order before preparation, and the temp-space preflight accounts for the rebuilt or joined image.
  • The descriptor is read more completely: medium type (CD/CD-R/CD-RW/DVD/DVD-R), each track’s extra block (pregap and length) and the footer data-file names (single-byte or UTF-16, *.mdf wildcards) are parsed, and the track mode now uses the low nibble of the mode byte exactly as libMirage’s reverse engineering established.
  • A CD descriptor whose sectors are the cooked 2048 bytes is converted through a MODE1/2048 cue instead of being treated as a DVD image.

ISZ support upgraded, library renamed to ISZSharp

  • Genuine UltraISO files now decode. The ISZ library was checked against libMirage’s ISZ filter and isz-tool, the two independent open-source readers, and now handles the real-file behaviours the published specification omits: the segment and chunk tables are de-obfuscated (XOR with B6 8C A5 DE, the complement of IsZ!) and bzip2 chunks get their stripped BZh header restored before decompression. Without those, a real .isz could not be decoded at all — the old test fixtures mirrored the same omission, so the suite passed while the reader could not open a genuine file.
  • More layouts are read: images whose header declares no chunk table (one raw run), and split images named game.part01.isz/game.part02.isz as well as the spec’s game.i01/game.i02.
  • UltraISO’s checksum is validated when the 64-byte header carries one; a mismatch is reported as damaged and the output deleted, like a size shortfall. A failed or cancelled decode now always deletes its partial output.
  • A split image whose cut lands inside a chunk now decodes. A later segment stores the tail of the straddling chunk (left_size) between its header and its chunk data; the reader started at the chunk data offset and skipped that tail, so a complete split image was rejected as truncated. Fixed in ISZSharp 1.0.1, and the split fixtures now use UltraISO’s real segment layout, so the round-trip tests exercise it.
  • The library now matches the other embedded packages’ shape: renamed UltraIsoSharp → ISZSharp, multi-targeting net8.0;net9.0;net10.0, shipping XML docs, a README and an icon, and published on NuGet (https://www.nuget.org/packages/ISZSharp) as 1.0.0, then 1.0.1 with the split-boundary fix.

Smaller fixes

  • Deleting originals for an Alcohol image now removes every data file the descriptor names and every volume of a split set, not just the first.
  • A multi-file .mds descriptor with one declared file missing is reported by name instead of silently falling back to the file that happens to be present, which could convert a truncated image.
  • A folder holding both .part1.rar and .part01.rar keeps the spelling whose next volume exists, so the set is still decodable.

Housekeeping

  • Version bumps: application 3.8.0, MDSSharp 1.1.0, ISZSharp 1.0.1.
  • Library updates: Meziantou.Analyzer 3.0.264.
  • Test suite grew to 896 tests (real multi-part RAR extraction, ISZ real-layout splits and checksums, MDS pregap/multi-file handling, RAR volume filtering).

3.7.2 (2026-09-16)

Recovered images with Mode 2 or subchannel layouts now convert (#67139)

  • A decoded ECM whose source was not a plain 2352-byte image is no longer reported as damaged. Alcohol .mdf rips can store 2336-byte Mode 2 sectors, or 2448/2368-byte sectors carrying subchannel data, and all three were skipped with “the decoded image is not a whole number of 2352-byte CD sectors or 2048-byte data sectors” even though the ECM trailing checksum had already proved the file intact. The new RecoveredImageClassifier routes every recovered image (ECM, ISZ, archive, split set) by its actual layout: raw 2352-byte CD sectors get a generated cue as before, 2048-byte sectors convert as a DVD image as before, 2336 and 2324-byte Mode 2 sectors get a MODE2/2336/MODE2/2324 cue (both round-trip losslessly through chdman 0.289), and 2448/2368-byte rips have their subchannel tail stripped to 2352 and are then sniffed and cued — the same strip the .mds path already used. Only a size that fits no standard layout is still reported as damaged.
  • The “not a whole number of 2352-byte CD sectors…” skip is now excluded from automatic bug reports, like the other user-data conditions.

Housekeeping

  • Test suite grew to 851 tests (recovered-image layout routing: 2336/2324 cues, 2448/2368 stripping, skip reasons).

3.7.1 (2026-09-16)

MDS descriptors find renamed and nested data files (#66955, #66989)

  • A .mdf no longer has to sit beside its .mds under the exact same name. The descriptor’s data file is now found when it was renamed with a decoration (Game.mds beside Game (USA).mdf — but never when the name continues with a letter or digit, so Game 2 never matches Game), when it sits one folder down (only an unambiguous exact-name match, so a sibling game’s image is never picked up), and split .i00 sets are located by base name even when other sets share the folder. File names that differ only in Unicode composition (Cafe\u0301 vs Café) are treated as equal.
  • Ambiguity is refused, not guessed: with Game (Disc 1).mdf and Game (Disc 2).mdf both beside Game.mds, the disc is skipped with the existing “data file was not found” message instead of converting the wrong disc.

Transient network/NAS failures no longer abort a batch (#66854)

  • Archive copies retry transient I/O errors before the temp-copy fallback runs: a failed direct extraction (for example an SMB hiccup while streaming from a NAS) is demoted to a debug notice, and the automatic copy that precedes the fallback retries up to four times with increasing delays. Missing-source errors are never retried.
  • Network-unavailability detection is locale-independent: the Win32 error codes (bad netpath, unexpected network error, netname deleted, …) are now read from the exception’s HResult, so non-English Windows builds (e.g. French "Erreur réseau inattendue.") are recognized without relying on localized message text.

Startup crash suppressed: desktop composition disabled (#67014, #67084)

  • A user with desktop composition disabled (registry/DWM tweak or third-party theming tool) got a COMException 0x80263001 from WindowChromeWorker.DwmExtendFrameIntoClientArea reported as an unhandled dispatcher exception. It is now suppressed like the other known-benign WPF-internal exceptions: the window simply runs without the glass frame effect.

Bug-report noise reduction (#67027)

  • “The output folder is not writable” (root of C:\, Program Files, a read-only drive) is excluded from automatic reports — the batch-start probe already shows one actionable dialog and nothing was converted.

Housekeeping

  • Release zips are leaner: the library .xml IntelliSense doc files (CCDSharp.xml, CSOSharp.xml, PBPSharp.xml) are no longer included — nothing at runtime uses them.
  • Library updates: Meziantou.Analyzer 3.0.259, Microsoft.NET.Test.Sdk 18.10.1.
  • Test suite grew to 842 tests (decorated/ambiguous/subdirectory/split/Unicode .mdf lookup, transient network copy retries, SFO size bounds, the new bug-report exclusion).

3.7.0 (2026-09-09)

Split archive volume sets now convert end to end

  • Extract split 7-Zip and ZIP sets in-app (.7z.001/.zip.001 style): what used to be skipped with “extract the set manually” is now extracted with the bundled 7za.exe and converted like any other input. The whole set is located from the first volume, disk space is checked against the total size of all volumes, and the extracted image is classified (cue/gdi/toc convert directly, bare images get the same recovery path as joined sets). Multi-part RAR still needs manual extraction, since the app does not carry RAR tooling.
  • Split-set skip notices (missing volume, no convertible image inside, missing 7za) are user-data conditions and are excluded from automatic bug reports.

Output folder convenience and resilience

  • Output folder auto-fill: selecting a source folder (browse or command line) now sets the Output CHD destination to the same folder when the output path is still empty. This is safe by construction — outputs are staged <name>.chdtmp and moved into place only on success, and existing CHDs are never inputs.
  • Disconnected output drive detected up front: a batch whose output folder sits on an unplugged USB stick or a dropped network drive now probes the folder once and reports a single actionable message, instead of failing every file with a confusing path error mid-batch.
  • Temp-folder cleanup is best effort: staged files carrying the read-only attribute (or locked by antivirus) no longer abort the batch or get reported as app bugs — attributes are cleared before deleting, directory read-only flags included, and a leftover temp folder is only logged as a warning.

PBP extraction — PBPSharp 1.1.0

The bundled PBP library was aligned with all four reference implementations (popstation, PSX2PSP, iPoPS, pop-fe), so PBP files that earlier versions rejected now extract:

  • pop-fe multi-disc PBPs no longer rejected: pop-fe leaves the four “template” DWORDs of the PSTITLEIMG000000 header zero (popstation/PSX2PSP write fixed magic values there). The header values are no longer validated — the disc position table, which is what actually locates the discs, is honoured regardless.
  • PARAM.SFO is no longer required: a missing or corrupt SFO leaves title/disc-id metadata empty instead of rejecting the whole file. None of the reference tools read the SFO when extracting disc images.
  • zlib-wrapped PSAR blocks decompress: blocks authored with zlib.compress (2-byte header + Adler-32) instead of raw deflate are retried as zlib streams after the raw inflate fails.
  • pop-fe index layout supported: the official 16-bit index entry (size uint16 at bytes 4–5, stored-flag byte at 6, SHA-1 at 8–23) is read directly; tools that write the size as a 32-bit int (popstation/PSX2PSP/iPoPS) are still covered because bytes 6–7 stay zero there.
  • Stored/uncompressed blocks flagged in the index (pop-fe --compression 0) are copied verbatim instead of being inflated.
  • Incompressible blocks accepted: index entries a few bytes larger than the raw 16-sector block (deflate framing on incompressible data) are no longer rejected.
  • 64-bit offset math for block reads, so multi-gigabyte files cannot overflow the 32-bit offset computation.
  • Hardened against corrupt indexes: an entry claiming to be a stored block larger than a full block is reported as corrupt data instead of overrunning the read buffer, and the split-set disk-space check now uses the whole volume set rather than the first volume alone.

Bug-report filtering additions

  • chdman crashes during the startup probe (entry-point-not-found / illegal-instruction exit codes — an installation incompatibility, not app logic) are excluded.
  • “Output folder is not available” notices (unplugged drive) are excluded.

Housekeeping

  • Version bumps: application 3.7.0, PBPSharp 1.1.0.
  • Library updates: Meziantou.Analyzer 3.0.231, Microsoft.NET.Test.Sdk 18.10.0.
  • Test suite grew to 835 tests (split-volume-set extraction, PBP index/deflate variants, pop-fe multi-disc, SFO tolerance, bug-report exclusions).

3.6.0 (2026-09-07)

Encoder resilience

  • A batch runs on the CHDSharp fallback when chdman is missing or fails its startup probe, instead of refusing to start. The startup dialog severity now reflects encoder priority (critical only when both encoders are unusable).

PBP extraction fixes

  • Corrupt deflate streams are classified as data errors: IndexOutOfRangeException leaking from SharpZipLib’s inflater is normalized to PbpError.DecompressionError, so a garbage PSAR block is reported as corrupt data instead of an app bug (regression-tested).
  • PBPSharp version bumps with the hardened block reader.

Shutdown & noise

  • Shutdown cancellation is treated as normal exit: cancelling the window during startup no longer reports a TaskCanceledException bug report.
  • Bug-report noise reduction: encoder-missing notices, chdman fallback notices, CHD open/read failures, file-move failures, encoder start failures and low-disk-space warnings are excluded from automatic reports (low disk space is now informational, since the conversion proceeds).
  • 0xC0000139 (entry point not found) crash decoding added alongside the existing illegal-instruction mapping.

Housekeeping

  • Library updates: NAudio 3.1.0, Meziantou.Analyzer 3.0.226; bundled 7-Zip binaries refreshed.
  • Application version bumped to 3.6.0; test suite at 820 tests.

This site uses Just the Docs, a documentation theme for Jekyll.