GeezerVison is showing github.com — Write a call argument as the caller wrote it in refinement errors by CatarinaGamboa · Pull Request #486 · liquid-java/liquidjava
Conversation
Problem
When a call argument is itself a call, the refinement error named its value after the callee's parameter (bound), which is not in the caller's code.
Before / after
Issue reproducer (return nextInt(size());):
before: bound³ >= 0 is not a subtype of bound³ > 0
Counterexample: bound³ == 0
after: size()³ >= 0 is not a subtype of size()³ > 0
Counterexample: size()³ == 0
Real case, fixture classes/found_binders_error (random.nextInt(items.size()), as in chaos-http-proxy's failures.size()):
before: bound⁶ == size(this.items) && bound⁶ >= 0 is not a subtype of bound⁶ > 0
∀bound⁶ bound⁶ == size(this.items) && bound⁶ >= 0
Counterexample: bound⁶ == 0
after: items.size()⁶ == size(this.items) && items.size()⁶ >= 0 is not a subtype of items.size()⁶ > 0
∀items.size()⁶ items.size()⁶ == size(this.items) && items.size()⁶ >= 0
Counterexample: items.size()⁶ == 0
Nested calls: nextInt(dec(size())) gives size()¹⁹ >= 0 && dec(size())²¹ == size()¹⁹ - 1 is not a subtype of dec(size())²¹ > 0. An unrefined call reads plain()⁸ == #plain_result⁷ (result name from #483).
The argument keeps its instance counter, like every other instance, so two calls written alike (between(next(), next())) stay apart.
Unchanged
Plain variables (positive(t)), field reads (bound == this.f), other expressions (bound == this.f - 1) and literals keep the parameter name: their refinement is bound == <their code>, so writing them as code would read as a tautology (0 == 0, which TestGeneratedNames would catch).
How
Same approach as FieldDisplayNames (#459): MethodsFunctionsChecker records the call's source text as written (Spoon's original source fragment, so count() is not printed as Basket.count()) for the argument instance in the new ArgumentDisplayNames. VariableFormatter.format writes it with the counter, and ExpressionNameResolver keeps it whole, as it does for fields (items.size() is not cut to size()). A call with no source position (built by Spoon, not read from a file) keeps the parameter name. CommandLineLauncher clears the record at the start of each run. The internal name and the SMT query are unchanged.
Size
+68 / −7 over 7 files (main code +48 / −2; the new ArgumentDisplayNames is 35 of those lines, most of them Javadoc).
Tests
TestFoundBinders: the reproducer's rows (also in the original step of the history) now use items.size(); new callArgumentIsWrittenAsTheCallerWroteIt on a new fixture line random.nextInt(count()), checking the message and counterexample read count(), not bound or Basket.count().
- All of them fail on
main, and the rows fail without the ExpressionNameResolver change.
mvn test: 693 tests, 0 failures.
Closes #481
🤖 Generated with Claude Code
Sorry, something went wrong.
Uh oh!
There was an error while loading. Please reload this page.
Codex review, round 1
- [P2]
Utils.java:190 — replaceAll("\\s+", " ") changes whitespace inside string literals and text blocks. A call such as check("a b") will appear in diagnostics as check("a b"), which is different source code. Collapse whitespace only outside literals, or preserve the source slice as written.
|
Sorry, something went wrong.
Uh oh!
There was an error while loading. Please reload this page.
Codex review, round 1: response
- Fixed: whitespace collapsing.
Utils.getSourceCode now only joins the lines of a multi-line argument (a line break and the indentation around it become one space), so spacing inside string literals is kept. A text block argument still has its lines joined, which is needed to show it on one line.
|
Sorry, something went wrong.
Uh oh!
There was an error while loading. Please reload this page.
Codex review, round 2
- [P2]
MethodsFunctionsChecker.java:468 — Distinct calls with identical source text receive the same display name. For example, two next() arguments can be separate values to the verifier but both appear as next() in a failed precondition or counterexample, producing misleading output such as next() < next(). Keep an occurrence marker or instance suffix when the display names collide.
|
Sorry, something went wrong.
Uh oh!
There was an error while loading. Please reload this page.
Codex review, round 2: response
- Fixed: calls written alike. The argument now keeps its instance counter like every other instance (
size()³ >= 0 is not a subtype of size()³ > 0), so between(next(), next()) shows next()⁴ and next()⁶. The PR description is updated with the new messages.
|
Sorry, something went wrong.
Uh oh!
There was an error while loading. Please reload this page.
Codex review, round 3
- [P2]
Utils.java:188 — Joining every source line also joins lines inside a Java text block passed to a call. The displayed binder and counterexample can therefore show different text from the argument the caller wrote. Preserve literal contents when formatting the snippet.
|
Sorry, something went wrong.
Uh oh!
There was an error while loading. Please reload this page.
Codex review, round 3: response
- Declined: text blocks inside a call argument. Message rows, binders and counterexamples are one line each, so a multi-line argument has to be joined to be shown at all; keeping a text block's line breaks would break the row layout. Only the line breaks change, never the characters on a line, and a text block passed to a refined call inside another call's argument is rare. No code change this round.
|
Sorry, something went wrong.
Uh oh!
There was an error while loading. Please reload this page.
Codex review, round 4
Nothing relevant found. The new argument names affect diagnostic formatting only; no change to the verifier's proof obligations and no concrete correctness issue.
|
Sorry, something went wrong.
Uh oh!
There was an error while loading. Please reload this page.
Codex review, round 5 (after rebasing on #483 and #484)
- [P2]
MethodsFunctionsChecker.java:465 — An invocation argument without a source position can crash verification here. Spoon returns a no-source fragment in that case, so reading its source code is unsafe. Check for a usable fragment and fall back to iArg.toString() for display.
|
Sorry, something went wrong.
Uh oh!
There was an error while loading. Please reload this page.
When an argument is a call, the error named its value after the callee's
parameter (bound), which is not in the caller's code. It is now written
as the call (items.size()) in Found, Expected and the counterexample,
keeping its instance counter. Only the display changes.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Codex review, round 5: response
- Fixed, differently: a call argument with no source position (one Spoon builds itself rather than reading from a file) is no longer renamed and keeps the parameter name. Falling back to
iArg.toString() would print Spoon's form (Basket.count()), which is not what the caller wrote either. Amended into the single commit.
|
Sorry, something went wrong.
Uh oh!
There was an error while loading. Please reload this page.
Uh oh!
There was an error while loading. Please reload this page.
Uh oh!
There was an error while loading. Please reload this page.
Assignees
No one assigned
Milestone
No milestone
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.
|