upgrade 'unevaluated array as argument' warning to error #305
Conversation
logger.warning does not deduplicate, in contrast to warnings.warn
|
Could you explain the context here? I'd imagine that this will be awfully noisy, for (IMO) not much gain. Ultimately, one should fix these warnings until they're all gone. |
My thought here was that in contrast to most (almost all?) other warnings, this type of warning can have serious performance implications. I was surprised to see this warning appear in the shock1d driver, and they appeared at a point during the execution where many other warnings are displayed, and thus this kind of warning tends to "disappear". I'll set this back to draft, I think we should probably also show the name of the compiled function. |
matthiasdiener
changed the title
You can set
If that's the case, wouldn't it be better to disable all warnings on a "production run"? |
TIL, thanks. The interface is a bit clumsy, but it seems to work.
Sorry, my text probably wasn't very clear here. What I meant is that the issue pointed out by the warnings could have serious implications. |
Oops, sorry, I wasn't clear there.. I meant that you can just filter warnings in your top-level driver, not here. I agree that it looks very clumsy as is :(
Ah, so the issue is that the warning is not loud enough? I would still argue that this is better done on the application side with |
arraycontext/impl/pytato/compile.py
Outdated
| "considerable expense. This is deprecated and will stop " | ||
| "working in 2023. To avoid this warning, force evaluation " | ||
| "of all arguments via freeze/thaw.", | ||
| stacklevel=4) |
There was a problem hiding this comment.
I think it'd be useful to create a special warning class for this:
class UnevaluatedCompiledFuncArgWarning(UserWarning):
pass
(and then pass the appropriate argument to warn) This makes precise filtering from the application side possible. I don't know that I want to install a warning filter on the user's behalf. (We could, in the actx constructor, on an opt-out basis? But that still feels overbearing.)
Alternatively, we could just make this an error, given that it has been deprecated with a deadline that has passed.
There was a problem hiding this comment.
I'm starting to like the "make it an error" idea, I think.
There was a problem hiding this comment.
I'm starting to like the "make it an error" idea, I think.
Same here :)
There was a problem hiding this comment.
Done in 5535398
matthiasdiener
changed the title
|
inducer/grudge#376 should address the grudge failures. |
| @@ -642,8 +631,10 @@ def __call__(self, arg_id_to_arg) -> ArrayContainer: | |||
| from .utils import get_cl_axes_from_pt_axes | |||
| from arraycontext.impl.pyopencl.taggable_cl_array import to_tagged_cl_array | |||
|
|
|||
| fn_name = self.pytato_program.program.entrypoint | |||
There was a problem hiding this comment.
This and the fn_names below appear to be inconsistent. Is there a specific reason for this?
There was a problem hiding this comment.
Thanks, this is hopefully fixed in b1c3a73. Part of the challenge here is that CompiledPyOpenCLFunctionReturningArray appears to be unused.
|
Thx! |
|
FYI @inducer @matthiasdiener, this change appears to break the non-compiled time integrators in grudge/mirgecom. (Looks like grudge downstream CI here doesn't run the examples, so it missed that.) |
As well it should! I'll get grudge fixed up. |
|
See inducer/grudge#377 and #307. |
Please squash