fix(LC0031): convert bare LockTable() to bare ReadIsolation in the code fix

[Arthurvdv]

Summary

Issue #530 reports LC0031 firing on bare LockTable() inside Base App table triggers (OnInsert, OnDelete) and notes that the code action "does not update anything here".

Verdict: not a false positive. The evidence:

1. SDK: the compiler binds bare, Rec., this. and named-variable receivers to one TableClassTypeSymbol built-in with no self-receiver special case; LockTable carries no deprecation (unlike FindSet(ForUpdate, UpdateKey) in the same class).
2. Microsoft Learn / BCQuality: LockTable is "a transaction-wide signal"; the endorsed use is a read directly preceding a dependent write. Neither mentions triggers as a special case.
3. BC App corpus (W1, BC 23.5 to 28.4): trigger LockTable() calls went from 80 to 78 (two removals, zero additions). Business Foundation, E-Document Core, Subscription Billing and Excise Taxes contain no LockTable at all.
4. Modern replacement: where Microsoft does need "make sure nobody took this number" in OnInsert, new code uses ReadIsolation on a second instance (Customer.Table.al, SustESGStandard.Table.al).

The real defect is the code fix returning the unchanged document on the bare form. The analyzer is unchanged.

What changed

• Code fix (RecordInstanceIsolationLevel.cs): add the IdentifierNameSyntax path so bare LockTable() becomes bare ReadIsolation(IsolationLevel::UpdLock). The MemberAccessExpressionSyntax path (named variable, Rec., this., RecordRef) keeps its receiver; any other shape still returns the document untouched. The replacement copies the trivia of the original invocation, because a fresh SyntaxFactory identifier would drop the indentation and comments of a bare call (found in review, pinned by HasFix/BareSelfWithLeadingComment).
• Fixtures: full receiver-form set (17 HasDiagnostic, 3 NoDiagnostic, 8 HasFix) pinning every receiver origin, RecordRef, arguments, and both ReadIsolation forms.
• Rule doc: new .claude/rules/diagnostics/lc0031-record-instance-isolation-level.md with the trigger decision, the per-form verdicts, SDK facts and code fix decisions; the receiver-forms regression row now covers the code fix flavour of the gap.

Receiver-form verdicts

Form	Analyzer	CodeFix
Named variable	ok, pinned	ok, pinned
Rec. (table, tableextension, page, TableNo OnRun)	ok, pinned	ok, pinned (table); same path elsewhere
Bare self in table and tableextension	ok, pinned	fixed, pinned
Bare self on page / in TableNo OnRun	ok, pinned	ok, pinned (page)
this. (table, tableextension)	ok, pinned (gated 14.0)	ok, pinned (gated 14.0)
Namespaced fully qualified variable	ok, pinned	same path as named variable
RecordRef variable	ok, pinned	ok, pinned
LockTable(true) (arguments)	ok, pinned	arguments dropped (known, documented)

Test plan

• HasFix/BareSelfInTrigger failed with an unchanged-document diff before the fix
• All 28 LC0031 tests pass after the fix (17 HasDiagnostic, 3 NoDiagnostic, 8 HasFix); tableextension fixtures are gated on SDK 13.0 (AL0334 on older compilers), this fixtures on runtime 14.0
• HasFix/BareSelfWithLeadingComment fails without WithTriviaFrom and passes with it
• Full LinterCop test suite: 384 passed, 2 skipped (pre-existing namespace skips)
• Three-TFM Release build (netstandard2.1;net8.0;net10.0): 0 warnings, 0 errors
• dotnet format ALCops.sln --verify-no-changes clean
• Validate-Rules.ps1: 54 rules files pass
• /code-review high (Sonnet): trivia loss on the bare form fixed; doc verdict table corrected; the suggested TryGetMethodCall reuse was tried and reverted because the helper returns the base ExpressionSyntax, which the factory methods reject
• CI green on all SDK legs (38 checks, Test results report green)

Closes #530
Related: #545

Docs companion PRs: ALCops/alcops.dev#185 (merged) and ALCops/alcops.dev#186 (tri-state version wording)

🤖 Generated with Claude Code