has_header doesn't return a bool on 0.52.0

[emersion]

With a Meson file containing the following:

if not cc.has_header('libavutil/hwcontext_drm.h', dependencies: libavutil)
	libavutil = disabler()
endif

Meson fails with "ERROR: Argument to "not" is not a boolean" when libavutil is not found (it's optional).

Downstream issue: swaywm/wlroots#1846

CI run with the failure: https://builds.sr.ht/~sircmpwn/job/94938#task-wlroots

CI run without the failure, Meson 0.51.2: https://builds.sr.ht/~sircmpwn/job/94937#task-wlroots

[xclaesse]

This seems to happen only when one of the dependencies is a disabler(). The minimal test case is

if not disabler()
endif

EDIT: As far as I know, it has always been an error, I dont think this is a regression. But this works:

if disabler()
endif

So there is definitely an inconsistency.

[emersion]

disabler() returns a dependency object, not a bool. However cc.has_header() returns a bool. So I don't think the if not disabler() case is related to this issue.

[xclaesse]

disabler() returns a dependency object, not a bool. However cc.has_header() returns a bool. So I don't think the if not disabler() case is related to this issue.

disabler() is a special beast. Any statement that contains a disabler object return a disabler. So has_header('foo.h', dependencies : disabler()) returns a disaber().

[xclaesse]

@emersion please give a try to #6011

[xclaesse]

@emersion I was currious how you're using disabler.

https://github.com/swaywm/wlroots/blob/master/examples/meson.build

• Add 'disabler' argument to functions returning not-found objects #3386 has been merged, you can remove the hack
• You don't need all_dep_found at all, disabler() is made exactly for that case. Unlike a not-found dependency, if you pass at least one disabler() to executable() then the whole function call is skipped. Not need of a if-block there.

[xclaesse]

@emersion I think I understood why it is a regression you're seeing. The problem is there:
https://github.com/swaywm/wlroots/blob/master/examples/meson.build#L19

set_variable('foo', disabler()) used to skip the set_variable() function call instead of actually setting the variable. That means that your dep variables where still not-found dependencies instead of disabler objects.

[emersion]

Thanks for the tips!

set_variable('foo', disabler()) used to skip the set_variable() function call instead of actually setting the variable. That means that your dep variables where still not-found dependencies instead of disabler objects.

Ah, indeed, that makes sense. And the if all_dep_found would handle that case later on.