meson introspect can hang in some cases

[kcgen]

Describe the bug
When fed DOSBox Staging's meson.build file, meson introspect hangs.

To Reproduce

cd /tmp
git clone --depth 1 https://github.com/dosbox-staging/dosbox-staging
cd dosbox-staging
meson introspect --projectinfo meson.build

Expected behavior
Printing of the project information to stdout, in JSON format.

system parameters

• cross build? : no
• operating system: Confirmed on Arch (latest), Ubuntu 22.04 and 23.04
• Python version: 3.11.2
• meson --version: 1.1.0

Thanks to @GranMinigun for flagging this via IRC.

[bruchar1]

The culprit is libsdlcd_sources = files(libsdlcd_sources) in src\libs\sdlcd\meson.build:20 .

Minimal example to reproduce:

project('sample', 'c')

f = ['sample.c']
f = files(f)

exe = executable('sample', f, install : true)

[bruchar1]

Similar problem occurs when performing a plusassignment. In ast/interpreter.py, assignment and evaluate_plusassign functions both replace the current assigned value by the new value, causing previous value to be lost. In the case of a +=, resulting introspection data would only preserve the added values of the assignment.

[bruchar1]

For example, meson introspect --targets meson.build on this file:

project('sample', 'c')

f = ['sample1.c']
f += ['sample2.c']

exe = executable('sample', f)

results in

[{"name": "sample", "id": "sample@exe", "type": "executable", "defined_in": "meson.build", "filename": ["sample.exe"], "build_by_default": true, "target_sources": [{"language": "unknown", "compiler": [], "parameters": [], "sources": ["C:\\env\\meson\\sample\\sample2.c"], "generated_sources": []}], "depends": [], "extra_files": [], "subproject": null, "installed": false}]

sample1.c is missing from introspection data.

[bruchar1]

Incidentally, the rewriter will work, because it will only alter the last assignment, but it can result in a duplicated source file because it doesn't know about previous assignments, and it will hang in an infinite loop with the first example.

[Volker-Weissmann]

Describe the bug When fed DOSBox Staging's meson.build file, meson introspect hangs.

To Reproduce

cd /tmp
git clone --depth 1 https://github.com/dosbox-staging/dosbox-staging
cd dosbox-staging
meson introspect --projectinfo meson.build

Expected behavior Printing of the project information to stdout, in JSON format.

system parameters

* cross build? : no

* operating system: Confirmed on Arch (latest), Ubuntu 22.04 and 23.04

* Python version: 3.11.2

* `meson --version`: 1.1.0

Thanks to @GranMinigun for flagging this via IRC.

If you replace meson.build with the path to a build directory it works:

meson introspect --projectinfo builddir

[GranMinigun]

This is an issue with IDEs like KDevelop and Qt Creator which do it that way.

[kcgen]

If you replace meson.build with the path to a build directory it works

@Volker-Weissmann , indeed, that's a workaround to side-step this bug.

The ticket's purpose is specifically to point out the bug exactly as specified in the reproduction steps to help the developers reproduce and fix it (not to work-around it).

[eli-schwartz]

Introspecting a build directory doesn't provide AST information about the meson.build files, so it isn't a workaround so much as doing something else entirely.

...

Minimal example to reproduce:

project('sample', 'c')

f = ['sample.c']
f = files(f)

exe = executable('sample', f, install : true)

poking around a bit:

meson/mesonbuild/ast/introspection.py

Lines 242 to 269 in a8b144b

	while inqueue:
	curr = inqueue.pop(0)
	arg_node = None
	assert isinstance(curr, BaseNode)
	if isinstance(curr, FunctionNode):
	arg_node = curr.args
	elif isinstance(curr, ArrayNode):
	arg_node = curr.args
	elif isinstance(curr, IdNode):
	# Try to resolve the ID and append the node to the queue
	assert isinstance(curr.value, str)
	var_name = curr.value
	if var_name in self.assignments:
	tmp_node = self.assignments[var_name]
	if isinstance(tmp_node, (ArrayNode, IdNode, FunctionNode)):
	inqueue += [tmp_node]
	elif isinstance(curr, ArithmeticNode):
	inqueue += [curr.left, curr.right]
	if arg_node is None:
	continue
	arg_nodes = arg_node.arguments.copy()
	# Pop the first element if the function is a build target function
	if isinstance(curr, FunctionNode) and curr.func_name in BUILD_TARGET_FUNCTIONS:
	arg_nodes.pop(0)
	elementary_nodes = [x for x in arg_nodes if isinstance(x, (str, StringNode))]
	inqueue += [x for x in arg_nodes if isinstance(x, (FunctionNode, ArrayNode, IdNode, ArithmeticNode))]
	if elementary_nodes:
	res += [curr]

files() is not an elementary node, so the value of f is appended to inqueue:

# third value of f, first non-assignment reference
IdNode(lineno=4, colno=10, filename='meson.build', end_lineno=4, end_colno=10)

Since this is an IdNode, we look up self.assignments[curr.value], and get:

FunctionNode(lineno=4, colno=4, filename='meson.build', end_lineno=4, end_colno=12)

This is where we get stuck, because the FunctionNode is pointing at the second assignment to f = , and has args.arguments with a value of the IdNode that we mentioned before... which is fair in context, but when we loop back around we look up the latest value of f via self.assignments.

Not sure what the good solution here is. One option might be to hoist some code out of resolve_node, which already has a loop detector.

[Volker-Weissmann]

To avoid duplicate work, please be informed that I'm currently working on it, fix will (hopefully) come this week.

[eli-schwartz]

Awesome, I wasn't working on it at all beyond the analysis I documented above.

[Volker-Weissmann]

While fixing this bug I found quite a lot of limitations/bugs in the meson rewriter. I think I'll rewrite some major parts of the rewriter.

[bruchar1]

@Volker-Weissmann did you finally fixed this bug?

[Volker-Weissmann]

@Volker-Weissmann did you finally fixed this bug?

Yes: It is waiting for review, then I will squash the commits and we can merge.

#12277