Inno Setup improvements - Separate script for languages and strings#36
Closed
bovirus wants to merge 2 commits into
Closed
Inno Setup improvements - Separate script for languages and strings#36bovirus wants to merge 2 commits into
bovirus wants to merge 2 commits into
Conversation
Contributor
|
@bovirus Both are in 1.9.2, which just shipped, thanks. The installer title now reads "InstallerClean 1.9.2" with no "versione", and the language and message definitions live in their own included file, as you suggested. On the title, you were right the first time: AppVerName was the answer. I'd reached for a NameAndVersion message override instead, but Inno Setup 6.7.1 doesn't recognise it (the compiler ignores it with a warning), so it did nothing; AppVerName=InstallerClean 1.9.2 is what actually drops the word, in both languages, for the title bar and the Add/Remove Programs entry. Inno Setup 7 drops it from the default too. Thanks for the nudge. So you know why your branch isn't merged directly: I rebuilt your contributions by hand against the repo rather than merging, because the files as sent would have bitten. The new languages file had no UTF-8 BOM but carries the accented Italian (è, à); without the BOM Inno reads it in the system ANSI codepage and garbles those accents, so I kept your structure with the BOM in place. They were also CRLF, where this repo is LF throughout. Thanks again. |
Author
|
Thanks to add AppVerName that solve the isuee one time for all. About new file probably could be because I uploaded it with a BOM marker different then the staandard on your Github page. |
Contributor
|
@bovirus No problem at all, that explains it. All sorted now. Thanks again for the AppVerName fix, that's the one that really nailed it. |
FarmLox
added a commit
that referenced
this pull request
Jul 19, 2026
Addresses every actionable finding from the independent ship-readiness
review at non-repo-files/v1.5.4-ship-readiness-review.md. Findings
that the review flagged but explicitly did not fix here are documented
inline with their reasoning.
Native interop (latent UB; harmless on x64 by ABI luck):
- SHFILEOPSTRUCT: Pack=8 -> Pack=1 to match the Windows SDK's
pshpack1.h declaration. Pack=8 inserted 4 bytes of padding before
pFrom on x64, putting the C# field at offset 16 while the kernel
reads from offset 12 - the shell happened to read field-aligned
bytes that just so happened to look right.
- MSIHANDLE: changed from IntPtr (8 bytes on x64) to uint (4 bytes,
matching `typedef unsigned long MSIHANDLE` in msi.h). The IntPtr
signature would crash on x86 where the calling convention pushes
8 bytes for a 4-byte argument. Cascaded the type change through
Msi.cs and MsiFileInfoService.cs.
- X509Certificate.CreateFromSignedFile result is now disposed too
(was leaked until finalisation; the X509Certificate2 wrapper
duplicates the handle so the inner needs its own using).
Operation correctness:
- F5 keybinding (and any other Re-scan trigger) now gates on
Cleanup.IsOperating and Completion.IsComplete via a new
ScanViewModel.IsExternallyBlocked flag flipped from MainViewModel.
Without this, F5 during a Move started a parallel scan that raced
the active operation.
- EnumeratePatches now throws InvalidOperationException after
MaxConsecutiveNonSuccess just like EnumerateProducts does. The
silent break would have left real-but-superseded patches missing
from the result set, classifying them as orphaned and offering
them for cleanup.
- EnumeratePatches now Array.Clear's its GUID buffers between
iterations, matching EnumerateProducts.
- Pre-flight write probe in CleanupViewModel.MoveAllAsync is now
cancellable: the operationCts is created BEFORE the probe (was
after), so the Cancel button on the operating overlay actually
cancels rather than waiting out the SMB timeout. The probe also
goes through the injected IFileSystem so MockFileSystem-driven
tests don't hit real disk.
- Settings file lost-update race fixed: SaveAfterDelayAsync re-loads
before saving, copies MoveDestination onto the freshly-loaded
AppSettings, then writes. Previously, the cached _settings instance
could clobber updates the detail-window code-behinds wrote
(window-size persistence) while the user was typing in the move-
destination textbox.
- PruneEmptySubdirectories called with CancellationToken.None: best-
effort cleanup, not an operation the user's Cancel should
re-classify a successful Move/Delete as cancelled.
- Move/DeleteFailed status pill now also includes the crash log path,
so users can find detail rather than guess where to look.
- Maximize/Restore button glyph now updates with the window state.
Path leakage under elevation:
- DescribeWriteFailure no longer routes ex.Message into the dialog.
Even an IOException's .Message can carry paths from outside the
user's typed destination (lock-holder process paths, NTFS
resolution chains); under elevation those could be paths from
another user's profile. Caller now writes to crash log first,
passes the log path through DescribeWriteFailure, and the dialog
body shows only the user's own dest plus the log-path pointer.
Resx patterns updated; matching tests inverted.
- ScanViewModel generic-exception catch shows a dialog (was status
pill only, easily missed).
- DispatcherUnhandledException in App.xaml.cs guards against re-
entry: a second exception during the MessageBox's nested message
pump no longer stacks two dialogs / two log entries.
CLI:
- Rejects extra positional args for /s, /d (silent truncation
before); /m allows args[1] only.
- /m PATH path with unquoted spaces ("/m D:\My Backup") used to
become "D:\My" with no warning.
- Three-state exit code: 0 success, 2 partial success, 1 full
failure. Documented in --help. Sysadmin retry policies can now
distinguish "all files failed" from "1 of 100 failed".
- DeletingFiles / MovingFiles status messages take a pluralised
noun parameter so "1 files" is no longer printed for single-
orphan runs.
- Console.CancelKeyPress handler registered BEFORE the mutex
acquisition so a Ctrl+C in the gap prints "Cancelling..."
gracefully.
- Pre-mutex-cancel path also unhooks the cancel handler before
return, matching the post-finally cleanup.
Threading / state:
- ScanViewModel re-samples HasPendingReboot AFTER the scan await,
not before. Sampling before could publish a stale "no pending
reboot" if Windows Update queued a reboot during a multi-second
scan, allowing Move/Delete to enable on stale state.
- PropertyChanged-based MoveDestination binding (was LostFocus): the
400 ms debounce now actually earns its keep (per-keystroke updates
let the placeholder hide on the first character and the Move
button enable as soon as the path becomes non-empty), and the
debounce prevents save thrashing.
- MoveAllAsync uses the captured `dest` consistently (was reading
MoveDestination live for confirmation / move call / restore-hint
string), closing a race where a fast user could change the
textbox between IsInstallerFolderOrChild validation and the Move
service call.
Localisation:
- Strings.en-GB.resx deleted: it was a 1:1 duplicate of the neutral
resx, and the project's NeutralLanguage is en-GB so .NET fallback
already returned the neutral resource for en-GB consumers. The
satellite assembly was loaded but its content was identical.
Maintenance trap eliminated.
- Resx patterns added: Cli.Help.ExitCodes* (the new --help block),
Cli.PendingRebootBlocked, Cli.EventLogPendingRebootBlocked.
UX polish:
- Result overlay tab-cycles inside the overlay's two buttons (was
cycling through main-window buttons behind the overlay until
focus eventually landed on Close).
- Result overlay has a 1px slate border so it reads as a layered
window instead of a free-floating block.
- Click on the dim margin around the result overlay dismisses it.
- Maximize/Restore button glyph swaps between U+25A1 and U+2750.
- Result-screen errors block is now keyboard-focusable so screen
readers can read it (overrides SelectableText style's
IsTabStop=False locally).
- AboutWindow Close button gets AutomationProperties.Name (was
inconsistent with the four other windows' close buttons).
- SubtleLink hyperlinks now underline on hover so the link is
discoverable without relying on colour shift alone.
Tests:
- DescribeWriteFailureTests: inverted to assert the inner exception
message does NOT leak and the log path DOES surface.
- MainViewModelTests timing-coupled debounce wait now reads
CleanupViewModel.MoveDestinationSaveDelay (made internal)
instead of hardcoding 700 ms.
- OrphanedFilesViewModelTests dispose-cancel test uses
TaskCompletionSource (await-based) instead of
DateTime.UtcNow polling.
- PendingRebootServiceTests removed the tautological
Assert.IsType<bool>(svc.HasPendingReboot()) test.
- PruneEmptySubdirectoriesTests gated on the
INSTALLERCLEAN_TEST_PRUNE=1 env var so they don't delete real
installer-folder subdirs on dev/CI hosts.
- InstallerClean.Tests.csproj sets RestorePackagesWithLockFile=true
so test dependency drift can't mask production behaviour.
Project notes drift:
- The notes said neutral resx is en-US; actually en-GB. Fixed and
noted the en-GB.resx satellite removal.
- The notes said only Core has SupportedOSPlatform; CLI has it too.
Now mentions both.
- "Nothing currently blocking" replaced with a pointer to the latest
ship-readiness review.
- MEMORY.md "no ex.Message to UI" claim qualified with the documented
inline exceptions (DescribeWriteFailure, ScanViewModel's
InvalidOperationException catch).
Findings explicitly NOT actioned (with reasoning):
- #18, #25, #59 are duplicates of #1, #11.
- #31 EventTrigger animation paths: reviewer's own conclusion was
"no fix unless profiler shows it".
- #32 Process.Start as Admin: would require explorer-token-drop
machinery; deferred (a click-once-per-session UX paper-cut).
- #36, #41, #44, #48, #53, #54, #57: reviewer confirmed not bugs.
- #40 RescanRequested via constructor: reviewer's "not a bug today".
- #43 AllowsTransparency: acceptable trade-off for confirm dialogs.
- #45, #46 manifest version: cosmetic.
- #47 Inno [Files] precondition: release-script concern, not
shipped product.
- #55, #56: minor UX nits.
- #58 CrashLog.Write return success/failure: would change the
signature across every caller; out of scope for a fixup pass.
178/178 tests passed before this commit; running on Windows after.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
@FarmLox
What this changes
Create a separate Inno Setup script that contains only languages and string definitions.
Adding custom NameAndVersion string for Italian to fix the issue regarding "versione" on top left.
This NameAndVersion string is fixed (without "versione" from Inno Setup 7.x - not version before).
The Inno Setup author confirmed that my feedfback is right ("versione" is not necessary) but apply this fix only starting from IS version 7.x.
Before IS 7,x the way to fix is adding custom string for NameAndVersion without "versione".
Why
Separate handling of language and strings by main script body