Fix api_list generation gaps - #592
Merged
Merged
Conversation
Five defects in the api_list generator, all of which caused shipping APIs to be missing or mis-described: * FeatureUnsafe was not in Identifiers.sharedIdentifiers, so InterlockedPolyfill.cs parsed to nothing and Interlocked.And/Or were absent, leaving an empty "#### Interlocked" section. Also added to the per-framework counter. * Properties were rendered without their type, so an entry read `HoursPerDay` rather than `long HoursPerDay`. * Properties in Polyfill_* files were never collected, so ProcessStartInfo.ArgumentList was missing entirely. * Members of an extension(Type) block in a Polyfill_* file were sectioned by their first parameter, scattering the six BitConverter statics across Double, Int32, Single, UInt32 and UInt64. * A //Link: or //Note: placed between a declaration's attributes and its modifiers was ignored, since an attribute list is part of the declaration and so is not in GetLeadingTrivia. That silently dropped the six BitConverter references, and with them the ?view=net-11.0 check that would have rejected those URLs. BuildApiTest and AssemblySizeTest both read Split, but nothing ordered them after SplitterTests.Run, so both could measure the previous run's output. That is why the per-framework counts were a run behind. Both now DependsOn it. API count 1088 -> 1091.
SimonCropp
added a commit
that referenced
this pull request
Sep 10, 2026
Covers both the non-generic and the generic TaskCompletionSource. On frameworks without the non-generic type the extension binds to the one Polyfill recreates, so a single implementation serves every target. Task does not expose the token that canceled it, so for a canceled source task the token is recovered from the OperationCanceledException the task throws when observed. Noted, since it costs a throw and catch the BCL avoids. Verified against net11: an incomplete task is rejected by TrySetFromTask with an ArgumentException rather than returning false, faulted inner exceptions are carried over by instance and are not flattened, and a canceled task with no token yields a token that cannot be canceled. Also fixes two api_list defects from #592: * Notes were emitted twice for members without an attribute list, since the declaration's leading trivia and its first modifier's leading trivia are the same trivia. Deduplicated by position, taking the note count from 107 to 61, which now matches the source exactly. * A recreated type listed with WriteType emitted a second header of the same name when the type also had extension members, as TaskCompletionSource now does. WriteType merges into the existing section; WriteHelper throws rather than writing members under a duplicate header. API count 1091 -> 1095.
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
1 participant
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.
Five defects in the api_list generator, all of which caused shipping APIs to be missing or mis-described.
FeatureUnsafewas not a known symbolIdentifiers.sharedIdentifierslisted every feature constant exceptFeatureUnsafe, soInterlockedPolyfill.cs(guarded#if !NET11_0_OR_GREATER && FeatureUnsafe && FeatureMemory) parsed to nothing under every moniker.Interlocked.AndandInterlocked.Orwere absent and#### Interlockedwas an empty section header. The same list is duplicated inCountApisForSplitFramework, so the per-framework counts under-counted them too.Properties rendered without their type
WriteSignature(Property)wrote only the identifier, so an entry read`HoursPerDay`where a method reads`TimeSpan FromDays(int, int)`. Now`long HoursPerDay`. This affects 22 existing entries.Properties in
Polyfill_*files were never collectedOnly static polyfill files had their properties read.
ProcessStartInfo.ArgumentList— a real shipping API with a//Link:— was missing from the list entirely.Extension-block members in
Polyfill_*files were sectioned by first parameterPolyfill_BitConverter.csuses anextension(BitConverter)block, so its methods have nothisparameter. The section was derived fromParameters[0], which scattered the six statics across#### Double,#### Int32,#### Single,#### UInt32and#### UInt64. Sectioning now comes from the enclosingextension(Type)receiver where there is one, falling back to the first parameter for classicthisextension methods.//Link:and//Note:after an attribute were ignoredAn attribute list is part of the declaration, so
member.GetLeadingTrivia()returns the trivia before the attribute — a comment sitting between[MethodImpl(...)]andpublic staticis leading trivia of thepublictoken instead, and was never seen. That silently dropped all six BitConverter references. It also meantLinkReader's?view=net-11.0check never ran on them, and all six URLs were indeed missing it.LinkReadernow scans the modifiers as well, and the six URLs are fixed.Ordering
BuildApiTest.RunWithRoslynandAssemblySizeTest.MeasureAssemblySizesboth readSplit, but nothing ordered them afterSplitterTests.Run. When they ran first they measured the previous run's Split output, which is why the committed per-framework counts were a run behind — the 15 net9TimeSpanconstants added in #591 never reached them. Both now[DependsOn]it, and this PR picks up that lag.Result
API count 1088 → 1091 (
Interlocked.And,Interlocked.Or,ProcessStartInfo.ArgumentList). Per-framework counts rise ~17 on pre-net9 targets, mostly the #591 lag.Encoding.Preamblealso had a fully-qualifiedSystem.ReadOnlySpan<byte>in the source, which rendered inconsistently; shortened.Reconciled every public member in
src/Polyfillagainst the regenerated list — the only remaining unlisted members are those of nested helper types (ChunkEnumerator,InvocationListEnumerator,SpanSplitEnumerator,SignalHelper), which is by design.Solution clean in Release, Consume clean across all 22 TFMs, tests green on net11.0 (1674), net10.0 (1674), net8.0 (1671), net462 (1640), plus PublicTests, EmbeddedTests, UnsafeTests, NoRefsTests and NoExtrasTests.