Enahancement/multifile max days

[Pr0metheus2]

This is a proposed fix for the Issue #735 and #185

Problem

The Max days setting only controlled how many previous calendar dates were searched when the MultiFile mask contained a $D date placeholder.

It did not check the last-modified date of files found using index patterns such as:

• *$J(.) → app.log, app.log.1, app.log.2
• *$J(.).log → app.log, app.1.log, app.2.log

As a result, all indexed rollover files were loaded regardless of their age.

Changes

• Added the file’s UTC last-write timestamp to ILogFileInfo.
• Implemented timestamp retrieval for:
  • Local files
  • SFTP files
• Applied the Max days limit when MultiFile files are discovered.
• The file explicitly opened by the user is always retained.
• MaxDayTry = 0 preserves the previous unlimited behavior.
• Filesystem plugins that cannot provide a timestamp retain the previous behavior and their files are not excluded.

Tests

Added coverage for both supported rollover filename layouts:

• app.log, app.log.1, app.log.2
• app.log, app.1.log, app.2.log

The tests verify that with MaxDayTry = 3:

• The active file is loaded.
• A file modified two days ago is loaded.
• A file modified four days ago is excluded.
• Setting MaxDayTry = 0 disables age filtering.

The existing $D date-placeholder rule remains working.

• added text hint (EN, DE CN)

[Hirogen]

hi, i added an analyses to the issue, its not really a bug

[Hirogen]

The PR doesn't do what the #735 spec asks. The spec wants a new, separate max-file-age option. The PR instead turns the existing MaxDayTry setting into an age limit. Because MaxDayTry still defaults to 3, every existing user with an index pattern like *$J(.) will silently lose rollover files older than 3 days. The spec lists exactly that change as out of scope.

Standards

Bugs

1. The "0 = unlimited" hint is wrong for $D date patterns. The date search still loops while (tryCounter < _options.MaxDayTry). With 0 that loop never runs, so no dated companion files load at all. The new hint text and the XML doc comment ("Zero disables the file-age limit") are only true for index patterns.
2. Existing users get a silent behaviour change. Index patterns had no age limit before. Now the default (and any saved MaxDayTry=3) drops app.log.N files older than 72 hours, and nothing tells the user.
3. Age is measured from now, not by calendar day. DateTime.UtcNow.AddDays(-MaxDayTry) is a rolling 72-hour window, but the date search counts calendar days. A file from day −3 written early in the morning is dropped even though the date search reached it.

Repo standards (judgement calls, none hard)

• Unrelated newline edits. A final newline is added to 9 .cs files the PR doesn't otherwise touch, and to Resources.zh-CN.resx. src/.editorconfig sets insert_final_newline = false for .cs, and .github/CLAUDE.md §3 says every changed line should trace to the request. The rule is a bit ambiguous, because the same editorconfig also sets resharper_csharp_insert_final_newline = true.
• Comment edits in MultiFileMaskDialog.Designer.cs. An existing // separator becomes //, and the new block doesn't match its neighbours. Same §3 rule.

Smells (judgement calls)

• One setting, two meanings. MaxDayTry now means both "missing days to skip" and "maximum file age", and its doc comment has to explain both.
• Duplicated code in SFTP. SftpLogFileInfo.LastWriteTimeUtc copies the try/_sftp.Get/catch shape of Length. It also adds an extra remote call for every file it checks.
• Over-broad catch. The SFTP getter catches ArgumentNullException, which Length next to it does not; catching it can hide bugs.
• Clock can't be faked. The code reads DateTime.UtcNow directly. The tests use real timestamps but leave a full day of margin, so they are stable enough.

Minor:

• TestData/app.log.2 ends with a trailing newline that app.log.1 doesn't have.
• The tests call Cleanup() inline, so a failing assert leaves the temp directory behind. The neighbouring tests do the same.
• The resource strings are consistent across all three languages and the designer file.

Spec (maintainer triage comment on #735)

Missing or partial

• No new option. The spec asks for "A new MultiFile option, max file age in days, with 0 meaning off". There's no new property; MaxDayTry is reused.
• Default isn't off. The spec says "0 is the default so existing behavior is unchanged". This is violated because the default stays at 3.
• No relabel. The spec asks to rename the label to something like "Max day gap" and to update the pattern-syntax help. The label still says "Max days:"; only a hint was added.
• Settings and Session File not touched. The spec asks for the option in Settings → MultiFile, and for a round-trip through SessionSnapshot, SessionFileComposer and PersistenceData, with old Session Files loading as 0. None of these files changed.
• No SDK version bump. The spec asks to "bump its own version in ColumnizerLib.csproj". The ILogFileInfo API changed, but the version didn't.
• Acceptance tests missing. There's no test for:
  • contiguity (an old app.log.1 followed by a new app.log.2)
  • the $D pattern being cut off by age
  • a file with no reported timestamp being included
  • the Session File round-trip
  • the setting being editable in both the dialog and Settings

Scope creep

• The spec lists "Changing what MaxDayTry does" as out of scope, and that is the core of the PR.

Implemented but wrong

• Clock not injectable. The spec says "Make 'now' injectable", but the code calls DateTime.UtcNow directly.
• Tests don't use fakes. The spec says tests should run "in Core against a fake file system and an injected clock, with no real file timestamps". The tests use real files, File.SetLastWriteTimeUtc and PluginRegistry.Instance.
• Stopping at the first old file only works for index patterns. The spec says the walk "stops at the first rollover file whose last-modified time is older". That holds for pure index patterns. In mixed patterns a too-old indexed file falls through to the date search, which keeps going. In date patterns a too-old file counts as a missing day.
• Done correctly: the new ILogFileInfo member defaults to null, a file with an unknown timestamp is included, and the local and SFTP versions are implemented.

Summary: 10 standards findings, the worst being the "0 = unlimited" hint, which actually loads no files for $D patterns. 12 spec findings, the worst being that the PR changes MaxDayTry instead of adding a separate option that defaults to off. The fix the spec points to is a new MaxFileAgeDays option (default 0) with an injected clock, carried through Settings and the Session File, plus a ColumnizerLib version bump.

[Pr0metheus2]

OK I will try to make your proposal as you require.

[Pr0metheus2]

Check, please

[Hirogen]

2 Bugs left:

• Overflow crash: a very large max file age crashes MultiFile loading.
• Language-dependent test: one UI test only passes on an English machine.

New:

• Bug: overflow crash. Both new up-downs allow values up to 2147483647. GetCandidateResult calls UtcDateTime.AddDays(-_options.MaxFileAgeDays), which throws ArgumentOutOfRangeException at about 740,000 days, and that breaks MultiFile loading. Fix: cap the up-downs (e.g. 3650) and/or clamp the value in code.
• Bug: test depends on UI language. src/LogExpert.UI.Tests/Dialogs/MultiFileAgeSettingsTests.cs:80 compares the label text, which is hard-coded in the Designer, against UIStrings.SettingsDialog_UI_Label_labelMaxDayGap. On a German or Chinese machine that fails. Fix: pin the culture, as the Highlight dialog tests do, or drop the assert.
• Judgement: unused resource keys. SettingsDialog never calls ApplyResources, so the three new SettingsDialog_UI_Label_* keys (in three languages) are never read; only the tooltips get localized. Either wire the labels up or don't add the keys.
• Judgement: nullable annotations. TimeProvider timeProvider = null should be TimeProvider?. TryGetFile(out SftpFile file) should be [NotNullWhen(true)] out SftpFile? file, like EncodingRegistry.TryGetEncoding.
• Smell: one concept, four names. MaxDayTry, MaxDayGap, MultiFileMaxDays and upDownMultifileDays all mean the same setting. The persisted names have to stay, so this is a judgement call.
• Smell: minor duplication. The two loops in GetNameList repeat the same Include/TooOld block. Tolerable at this size.
• Minor: SFTP cache. _lastWriteTimeUtc isn't cleared when TryGetFile fails. That's harmless for the short-lived info objects the rollover search uses.

Remaining:

• No test for global-default seeding. The spec says "the global default seeds new windows". The code does this at LogWindow.cs:2851 and :2901, but no test asserts it.
• Scope creep: SFTP error handling changed. With the shared TryGetFile, FileExists no longer catches ArgumentNullException and Length now also catches SshException. It's small and arguably an improvement.
• Scope creep: XML read of a new field. PersisterXML reads a maxFileAgeDays attribute that no XML writer ever produces. It's harmless dead code.