Pharo test defects

StDebugger-skips

State
Fixed (image fix)
Defect in
the test
Tests
9
Fix
image fix 20-StDebugger-skips.st
Upstream repository
pharo-spec/NewTools

Record

All 9 StDebugger skips are now real passing tests on both VMs. No VM defect was found. Defects: 2 obsolete expectations ('0' in the StDebuggerContextPredicate printDescription tests); 1 skip with no remaining reason (testComputeInitialTopContext); testGenerateAndProceed (test harness re-ran the test: the debugger session wraps a context whose sender chain is the test's own chain, so resuming it re-executes the test; plus CR vs LF in expected sources and an obsolete 'see:' expectation; the test also never checked the recompiled method, now it does); 5 empty DBG_TEST placeholders (now real tests of StDebuggerInspector save/restoreRawInspectionSelectionForContext: and StDebugger>>returnEnteredValue). All repairs are in the tests only; no image code under test was changed. Side findings not fixed: (a) StDebuggerInspector>>restoreRawInspectionSelectionForContext: has 'path ifEmpty: [ self rawInspectionSelectionCache at: receiverClass ifAbsent: [ #() ] ]' whose result is discarded, so the receiver-class-only fallback never takes effect (probable upstream bug; not changed because it alters debugger behaviour); (b) selecting the outer-temp nodes 2-4 of the session fixture in the raw inspector fails with SubscriptOutOfBounds when the home method context of StTestDebuggerProvider>>sessionForDebuggerTests is widowed (sp=0), which happens on both VMs from the second call of that method in an image on Cog (first call keeps sp=3), so the new tests select only paths 1 and 5.

Test Defect in State Cause
StDebuggerContextPredicateTest>>testPrintDescription test repaired Skipped and expected '0'; StDebuggerContextPredicate>>printDescription answers the post-mortem prefix plus 'context printString' (source read in the image). Expectation changed to context printString.
StDebuggerContextPredicateTest>>testPrintPostMortemDescription test repaired Same obsolete '0'; printPostMortemDescriptionOn: writes '[Post-mortem] ' then printDescription adds context printString. Expectation changed accordingly.
StDebuggerClientModelTest>>testComputeInitialTopContext test repaired Skipped with the remark that it makes production CI fail. With the skip removed and the body unchanged it passes on both VMs.
StDebuggerTest>>testGenerateAndProceed test repaired Resuming the DebugSession resumes the sender chain of the test itself in a new process, so the rest of the test runs a second time (measured: DNU #application: on StTestDebuggerProvider, then tearDown DNU). Debugger context is now created in a forked process. Also expected sources used LF while image method sources use CR, and the expected generated code was the obsolete 'see: ... equals: 111' instead of the 'assert: ... equals: 111' that StGenerateAssertEqualsCode>>generatedCodeFromDebugger: writes. The test now also asserts the recompiled method source, which it computed but never used.
StDebuggerTest>>testSaveReceiverRawInspectionSelection test repaired Empty placeholder (self skip. self flag: #DBG_TEST). Now selects path #(1) in the raw inspector, calls saveRawInspectionSelectionForContext: and asserts both cache keys (class->selector and class) hold #(1).
StDebuggerTest>>testSaveReceiverRawInspectionSelectionAfterContextChange test repaired Empty placeholder. Now steps once, selects path #(5), steps into another context and asserts the old context's selection was saved in the cache. Paths 2-4 avoided because they read temps of the returned (widowed) home context of the fixture method.
StDebuggerTest>>testRestoreReceiverRawInspectionSelection test repaired Empty placeholder. Now puts #(5) in the cache for the context's receiver class->selector, calls restoreRawInspectionSelectionForContext: and asserts the raw inspector selected path is #(5).
StDebuggerTest>>testRestoreReceiverRawInspectionSelectionAfterContextChange test repaired Empty placeholder. Now selects #(5) in the first context, steps into another context (selection no longer #(5)), selects the first stack item again and asserts #(5) is restored.
StDebuggerTest>>testReturnEnteredValue test repaired Empty placeholder. StDebugger>>returnEnteredValue opens a modal request dialog, so the test gives the debugger an anonymous StPharoApplication subclass whose newRequest answers a stub request: first nil (user cancels: interruptedContext unchanged, the case the DBG_HowToTestNilIfUserCancels flag asks for), then '0' (the squared context is popped, interruptedContext is its sender and its top is 0).

Regression check: Only test methods were changed (no non-test method), so the covering classes are the owner classes plus all 20 StDebugger*/StGenerate*Test classes (StDebuggerAssertionContextTest, ClientModelAnnouncementTest, ClientModelTest, CommandTest, CommandTreeBuilderTest, ContextInteractionModelTest, ContextPredicateTest, ContextTest, ExceptionExtensionsTest, ExtensionMechanismTest, InspectorTest, MNUExtensionTest, RawObjectInspectorTest, ToolCommandTreeBuilderTest, TreeTablePresenterTest, StDebuggerTest, and the four *CommandTreeBuilderTest classes). Totals of PASS over these classes, without fix vs with fix: ours VM 284 -> 294 (non-pass 10 -> 0); Cog (arm64 pharo-vm via runtest.sh cog) 285 -> 294 (non-pass 9 -> 0; the 9 are the 9 skips, the old 10th on ours/9th... base Cog also had testUpdateExtensionSubscription fail once in an earlier run, not in this run). No test that passed before failed after. Repeat runs: the 6 StDebuggerTest tests ran together 4 times on each VM, all pass; the 3 other tests ran 3 times on each VM, all pass. Earlier intermediate runs with an earlier version of the tests (selecting paths 2/3) failed intermittently with SubscriptOutOfBounds on both VMs; that version is replaced.

Upstream note: Pharo NewTools (StDebugger tests): un-skip the StDebugger tests. (1) StDebuggerContextPredicateTest: expect context printString (with the '[Post-mortem] ' prefix), not '0'. (2) StDebuggerClientModelTest>>testComputeInitialTopContext: remove skip, passes. (3) StDebuggerTest>>testGenerateAndProceed: create the generated-code debugger in its own process (resuming the session otherwise re-runs the test method), compare sources modulo line ends, expect 'assert: ... equals: 111' and assert the recompiled method. (4) Replace the five #DBG_TEST placeholders with tests of StDebuggerInspector>>save/restoreRawInspectionSelectionForContext: and StDebugger>>returnEnteredValue (using a stub application whose newRequest answers the expression). Possible upstream bug noticed, not changed here: StDebuggerInspector>>restoreRawInspectionSelectionForContext: discards the result of 'path ifEmpty: [ cache at: receiverClass ifAbsent: [ #() ] ]'. These changes need review by the owner before any PR.

Results

Last full sweeps of 2026-11-17: iospharo (sweep14-new14) and stock Cog with the Cog changes (sweepcogp14-new10), with all image fixes. "Before" is the state of the test before the repairs. The Stock Cog column has a value only where the record measured stock Cog: "PASS 3 of 3 (repair run)" is the check of each repair on stock Cog v12.0.6 with the fix entry; "-" is not measured separately.

TestBeforeStateiospharo VMPatched CogStock Cog
StDebuggerClientModelTest>>testComputeInitialTopContextSKIPrepairedPASSPASSPASS 3 of 3 (repair run)
StDebuggerContextPredicateTest>>testPrintDescriptionSKIPrepairedPASSPASSPASS 3 of 3 (repair run)
StDebuggerContextPredicateTest>>testPrintPostMortemDescriptionSKIPrepairedPASSPASSPASS 3 of 3 (repair run)
StDebuggerTest>>testGenerateAndProceedSKIPrepairedPASSPASSPASS 3 of 3 (repair run)
StDebuggerTest>>testRestoreReceiverRawInspectionSelectionSKIPrepairedPASSPASSPASS 3 of 3 (repair run)
StDebuggerTest>>testRestoreReceiverRawInspectionSelectionAfterContextChangeSKIPrepairedPASSPASSPASS 3 of 3 (repair run)
StDebuggerTest>>testReturnEnteredValueSKIPrepairedPASSPASSPASS 3 of 3 (repair run)
StDebuggerTest>>testSaveReceiverRawInspectionSelectionSKIPrepairedPASSPASSPASS 3 of 3 (repair run)
StDebuggerTest>>testSaveReceiverRawInspectionSelectionAfterContextChangeSKIPrepairedPASSPASSPASS 3 of 3 (repair run)

Test

Run the tests in a Pharo 14 image (build 791). Before the fix they skip, fail or carry the expected-failure mark.

"Run in a Playground of a Pharo 14 image (Do it and print each line)."
((StDebuggerClientModelTest selector: #testComputeInitialTopContext) run) printString.
((StDebuggerContextPredicateTest selector: #testPrintDescription) run) printString.
((StDebuggerContextPredicateTest selector: #testPrintPostMortemDescription) run) printString.
((StDebuggerTest selector: #testGenerateAndProceed) run) printString.
((StDebuggerTest selector: #testRestoreReceiverRawInspectionSelection) run) printString.
((StDebuggerTest selector: #testRestoreReceiverRawInspectionSelectionAfterContextChange) run) printString.
((StDebuggerTest selector: #testReturnEnteredValue) run) printString.
((StDebuggerTest selector: #testSaveReceiverRawInspectionSelection) run) printString.
((StDebuggerTest selector: #testSaveReceiverRawInspectionSelectionAfterContextChange) run) printString.

Source of the tests before the fix

StDebuggerClientModelTest>>testComputeInitialTopContext before the fix (Pharo 14 build 791)
testComputeInitialTopContext

	| newSession dummyActionModel |
	self flag:
		'We suspect this test to make the production CI fail so we skip it for now'.
	self skip.
	dummyActionModel := StTestDebuggerProvider new
		                    debuggerWithMissingSubclassResponsibilityContext
		                    debuggerClientModel.
	self changeSession: dummyActionModel session.
	dummyActionModel clear.
	self
		assert: debugActionModel topContext
		identicalTo: session exception signalContext.

	dummyActionModel := StTestDebuggerProvider new
		                    debuggerWithMissingSubclassResponsibilityContext
		                    debuggerClientModel.
	newSession := dummyActionModel session.
	dummyActionModel clear.
	newSession restart: newSession interruptedContext sender.
	self changeSession: newSession.
	self
		assert: debugActionModel topContext
		identicalTo: session interruptedContext
StDebuggerContextPredicateTest>>testPrintDescription before the fix (Pharo 14 build 791)
testPrintDescription
	self skip.
	self assert: predicate printDescription equals: '0'
StDebuggerContextPredicateTest>>testPrintPostMortemDescription before the fix (Pharo 14 build 791)
testPrintPostMortemDescription
	self skip.
	predicate postMortem: true.
	self assert: predicate printDescription equals: '[Post-mortem] 0'
StDebuggerTest>>testGenerateAndProceed before the fix (Pharo 14 build 791)
testGenerateAndProceed

	| baseSourceCode expectedSourceCode adapter |
	self skip:
		'Does not work, I do not know why it looks like my test is executed several times with the same debugger'.
	baseSourceCode := 'methodThatGeneratesCode

	| asserter |
	asserter := TestAsserter new.
	
	asserter see: 42 + 66 + 3'.

	expectedSourceCode := 'methodThatGeneratesCode

	| asserter |
	asserter := TestAsserter new.
	
	asserter see: 42 + 66 + 3 equals: 111'.

	debugger := StTestDebuggerProvider new
		            debuggerWithGeneratingCodeContext
		            application: self debuggerClass currentApplication;
		            initialize;
		            yourself.
	"just things necessary to use debugger's code presenter, which we need in this test"
	adapter := SpMorphicCodeAdapter new.
	adapter adapt: debugger code.
	debugger code adapter: adapter.
	debugger code adapter buildWidget.

	self assert: debugger canGenerateAndProceed.
	self assert: debugger exception isGeneratingCode.
	self
		assert: debugger interruptedContext sender sourceCode
		equals: baseSourceCode.

	oldMethodToSourceDictionary
		at: debugger interruptedContext sender method
		put: baseSourceCode.

	debugger saveGeneratedCodeAndProceed
	"self assert: debugger interruptedContext isNil"
StDebuggerTest>>testRestoreReceiverRawInspectionSelection before the fix (Pharo 14 build 791)
testRestoreReceiverRawInspectionSelection
	self skip.
	self flag: #DBG_TEST
StDebuggerTest>>testRestoreReceiverRawInspectionSelectionAfterContextChange before the fix (Pharo 14 build 791)
testRestoreReceiverRawInspectionSelectionAfterContextChange

	self skip.
	self flag: #DBG_TEST
StDebuggerTest>>testReturnEnteredValue before the fix (Pharo 14 build 791)
testReturnEnteredValue
	self skip.
	self flag: #DBG_TEST
StDebuggerTest>>testSaveReceiverRawInspectionSelection before the fix (Pharo 14 build 791)
testSaveReceiverRawInspectionSelection

	self skip.
	self flag: #DBG_TEST
StDebuggerTest>>testSaveReceiverRawInspectionSelectionAfterContextChange before the fix (Pharo 14 build 791)
testSaveReceiverRawInspectionSelectionAfterContextChange

	self skip.
	self flag: #DBG_TEST

Fix

How to apply an entries file: file in 00-header.st and the entries file as one chunk (the iospharo script scripts/gen-image-fixes.py makes that chunk). Each entry checks for the defect first and prints FIXED, ADDED or SKIP. Download 00-header.st.

20-StDebugger-skips.st (download)
"StDebuggerContextPredicateTest>>testPrintDescription and testPrintPostMortemDescription
skipped themselves and expected '0'.  StDebuggerContextPredicate>>printDescription
answers the post-mortem prefix (from printPostMortemDescriptionOn:) followed by
'context printString'.  The '0' is a leftover from an older printDescription that
printed a number.  The tests now expect what printDescription is written to answer."
fix value: #StDebuggerContextPredicateTest value: #testPrintDescription
	value: [ :source | source includesSubstring: 'self skip' ]
	value:
'testPrintDescription

	self assert: predicate printDescription equals: context printString'.


fix value: #StDebuggerContextPredicateTest value: #testPrintPostMortemDescription
	value: [ :source | source includesSubstring: 'self skip' ]
	value:
'testPrintPostMortemDescription

	predicate postMortem: true.
	self
		assert: predicate printDescription
		equals: ''[Post-mortem] '' , context printString'.

"StDebuggerClientModelTest>>testComputeInitialTopContext was skipped with the remark
that it makes the production CI fail.  It passes on both VMs; skip removed."
fix value: #StDebuggerClientModelTest value: #testComputeInitialTopContext
	value: [ :source | source includesSubstring: 'self skip' ]
	value:
'testComputeInitialTopContext

	| newSession dummyActionModel |
	dummyActionModel := StTestDebuggerProvider new
		                    debuggerWithMissingSubclassResponsibilityContext
		                    debuggerClientModel.
	self changeSession: dummyActionModel session.
	dummyActionModel clear.
	self
		assert: debugActionModel topContext
		identicalTo: session exception signalContext.

	dummyActionModel := StTestDebuggerProvider new
		                    debuggerWithMissingSubclassResponsibilityContext
		                    debuggerClientModel.
	newSession := dummyActionModel session.
	dummyActionModel clear.
	newSession restart: newSession interruptedContext sender.
	self changeSession: newSession.
	self
		assert: debugActionModel topContext
		identicalTo: session interruptedContext'.

"StDebuggerTest>>testGenerateAndProceed was skipped: 'it looks like my test is
executed several times with the same debugger'.  That is what happens: the
debugger session wraps the context that signalled StGenerateCode, and the
sender chain of that context is the chain of the test itself.  Proceeding the
session resumes the chain in a new process, which runs the rest of the test
method again (measured: the second run fails with 'StTestDebuggerProvider does
not understand #application:' and tearDown fails on the replaced debugger).
The test now creates the context in a process of its own.  Two more repairs:
the method source of the image has CR line ends and the expected sources in the
test had LF, and the expected generated code is the assert:equals: send that
StGenerateAssertEqualsCode>>generatedCodeFromDebugger: writes in place of see:
(the old expectation 'see: ... equals: 111' was never produced).  The test
also asserted nothing after proceeding although it computed
expectedSourceCode; it now checks the recompiled method."
fix value: #StDebuggerTest value: #testGenerateAndProceed
	value: [ :source | source includesSubstring: 'self skip' ]
	value:
'testGenerateAndProceed

	| baseSourceCode expectedSourceCode adapter provider created createdSignal normalized |
	"The image keeps method sources with CR line ends; compare without that difference."
	normalized := [ :text | text withLineEndings: String cr ].
	baseSourceCode := ''methodThatGeneratesCode

	| asserter |
	asserter := TestAsserter new.
	
	asserter see: 42 + 66 + 3''.

	expectedSourceCode := ''methodThatGeneratesCode

	| asserter |
	asserter := TestAsserter new.
	
	asserter assert: 42 + 66 + 3 equals: 111''.

	"The debugger session wraps the context that signalled StGenerateCode.  Its
	sender chain must not be the chain of this test: proceeding the session
	resumes that chain in a new process, which would run the rest of this test
	method a second time (this is why the test used to be skipped).  So the
	context is created in a process of its own, and this method keeps the
	debugger in its own variable, which the resumed chain cannot reach."
	provider := StTestDebuggerProvider new.
	createdSignal := Semaphore new.
	[ created := provider debuggerWithGeneratingCodeContext.
	createdSignal signal ] forkAt: Processor userSchedulingPriority.
	(createdSignal wait: 10 seconds) ifTrue: [ self fail: ''debugger was not created'' ].
	debugger := created.
	debugger
		application: self debuggerClass currentApplication;
		initialize.
	"just things necessary to use debugger''s code presenter, which we need in this test"
	adapter := SpMorphicCodeAdapter new.
	adapter adapt: debugger code.
	debugger code adapter: adapter.
	debugger code adapter buildWidget.

	self assert: debugger canGenerateAndProceed.
	self assert: debugger exception isGeneratingCode.
	self
		assert: (normalized value: debugger interruptedContext sender sourceCode)
		equals: (normalized value: baseSourceCode).

	oldMethodToSourceDictionary
		at: debugger interruptedContext sender method
		put: baseSourceCode.

	debugger saveGeneratedCodeAndProceed.
	"StGenerateAssertEqualsCode>>generatedCodeFromDebugger: replaces the see: send
	by assert:...equals: with the value that was seen."
	self
		assert: (normalized value: (StTestDebuggerProvider >> #methodThatGeneratesCode) sourceCode)
		equals: (normalized value: expectedSourceCode)'.

"The five StDebuggerTest tests below were empty placeholders (self skip. self flag:
#DBG_TEST).  They now test StDebuggerInspector>>saveRawInspectionSelectionForContext:
and restoreRawInspectionSelectionForContext:, which no other test covers."
fix value: #StDebuggerTest value: #testSaveReceiverRawInspectionSelection
	value: [ :source | source includesSubstring: 'self skip' ]
	value:
'testSaveReceiverRawInspectionSelection

	| context |
	"Paths 1 (self) and 5 (the temporary i of the block) can be read in the context of
	the session; the nodes 2 to 4 are temporaries of the method that created the
	session, which has returned, and a returned context cannot be read."
	dbg := self debuggerOn: session.
	context := dbg currentContext.

	dbg inspector getRawInspectorPresenterOrNil selectPath: #( 1 ).
	dbg inspector saveRawInspectionSelectionForContext: context.

	self
		assert: (dbg inspector rawInspectionSelectionCache
				 at: context receiver class -> context selector)
		equals: #( 1 ).
	self
		assert:
		(dbg inspector rawInspectionSelectionCache at: context receiver class)
		equals: #( 1 )'.


fix value: #StDebuggerTest value: #testSaveReceiverRawInspectionSelectionAfterContextChange
	value: [ :source | source includesSubstring: 'self skip' ]
	value:
'testSaveReceiverRawInspectionSelectionAfterContextChange

	| context |
	dbg := self debuggerOn: session.
	dbg stepInto. "Still in the same context, the raw inspector is updated"
	context := dbg currentContext.

	dbg inspector getRawInspectorPresenterOrNil selectPath: #( 5 ).
	dbg stepInto. "We enter another context: the selection of the old one is saved"

	self deny: dbg currentContext identicalTo: context.
	self
		assert: (dbg inspector rawInspectionSelectionCache
				 at: context receiver class -> context selector)
		equals: #( 5 )'.


fix value: #StDebuggerTest value: #testRestoreReceiverRawInspectionSelection
	value: [ :source | source includesSubstring: 'self skip' ]
	value:
'testRestoreReceiverRawInspectionSelection

	| context |
	dbg := self debuggerOn: session.
	context := dbg currentContext.
	self assert: dbg inspector getRawInspectorSelectedPath equals: #(  ).

	dbg inspector rawInspectionSelectionCache
		at: context receiver class -> context selector
		put: #( 5 ).
	dbg inspector restoreRawInspectionSelectionForContext: context.

	self assert: dbg inspector getRawInspectorSelectedPath equals: #( 5 )'.


fix value: #StDebuggerTest value: #testRestoreReceiverRawInspectionSelectionAfterContextChange
	value: [ :source | source includesSubstring: 'self skip' ]
	value:
'testRestoreReceiverRawInspectionSelectionAfterContextChange

	| firstContext |
	dbg := self debuggerOn: session.
	dbg stepInto. "Still in the same context, the raw inspector is updated"
	firstContext := dbg currentContext.
	dbg inspector getRawInspectorPresenterOrNil selectPath: #( 5 ).

	dbg stepInto. "We enter another context"
	self deny: dbg currentContext identicalTo: firstContext.
	self deny: dbg inspector getRawInspectorSelectedPath equals: #( 5 ).

	"Back to the first context: its selection is restored"
	dbg stackTable selectItem: dbg stackTable items last.
	self assert: dbg currentContext identicalTo: firstContext.
	self assert: dbg inspector getRawInspectorSelectedPath equals: #( 5 )'.

"StDebuggerTest>>testReturnEnteredValue: StDebugger>>returnEnteredValue asks the user for an
expression with a modal request dialog.  The test gives the debugger an
application whose request answers the expression '0' and, in a second run,
nil (the user cancelled), the case that the flag DBG_HowToTestNilIfUserCancels
in returnEnteredValue asks for."
fix value: #StDebuggerTest value: #testReturnEnteredValue
	value: [ :source | source includesSubstring: 'self skip' ]
	value:
'testReturnEnteredValue

	| requestClass applicationClass application context sender |
	requestClass := Object newAnonymousSubclass.
	requestClass addInstVarNamed: ''answer''.
	requestClass compile: ''answer: anObject answer := anObject''.
	requestClass compile: ''title: aString ^ self''.
	requestClass compile: ''openModal ^ answer''.
	applicationClass := StPharoApplication newAnonymousSubclass.
	applicationClass addInstVarNamed: ''request''.
	applicationClass compile: ''request: anObject request := anObject''.
	applicationClass compile: ''newRequest ^ request''.
	application := applicationClass new.

	dbg := self debugger
		       session: session;
		       application: application;
		       initialize;
		       yourself.
	dbg stepInto; stepInto. "We enter the squared method"
	context := dbg interruptedContext.
	sender := context sender. "a returned context has no sender any more"

	"The user cancels the request: nothing is returned"
	application request: (requestClass new answer: nil; yourself).
	dbg returnEnteredValue.
	self assert: dbg interruptedContext identicalTo: context.

	"The user enters an expression: its value is returned to the sender"
	application request: (requestClass new answer: ''0''; yourself).
	dbg returnEnteredValue.
	self assert: dbg interruptedContext identicalTo: sender.
	self assert: dbg interruptedContext top equals: 0'.

State of 2026-11-20. Made from the test records of the iospharo project (a second Pharo VM). The fixes are proposals for review.