Improve WebP encoder error handling

[akx]

Follows up on #5471 and #9993.

Instead of reporting just "error 6", this PR adds human-readable error messages for the WebP encoder errors. The error messages are based on the WebP header.

The memory-hungry (675 MB allocated for test image) test_write_encoding_error_message test case (that tested "partition is bigger than 512k") is no longer necessary, since test_write_encoding_error_bad_dimension covers the same code path.

That error message is improved too, though:

>>> from PIL import Image
>>> Image.new("RGB", (15000, 15000)).save("foo.webp", method=0)
[...]
ValueError: encoding error 6: partition #0 is bigger than 512K

[akx]

See https://github.com/webmproject/libwebp/blob/13977a505203ae4e3641ce71b6a60738ad66c564/src/webp/encode.h#L284-L297

[radarhere]

I'm reluctant to remove the test, since it is guarding against regression for the specific case of #5461.

[radarhere]

What is the story with WEBP_STR and WEBP_STR_HELPER? Why do they need to be defined separately, and why are there two of them?

[akx]

That's just C preprocessor expansion rules and standard procedure:

a.c

#define WEBP_MAX_DIMENSION 123

// Doesn't work like we want
#define WEBP_STR_1(x) #x
WEBP_STR_1(WEBP_MAX_DIMENSION)

// Works like we want
#define WEBP_STR_HELPER(x) #x
#define WEBP_STR_2(x) WEBP_STR_HELPER(x)
WEBP_STR_2(WEBP_MAX_DIMENSION)

Preprocessing:

$ gcc -E a.c
# 1 "a.c"
# 1 "<built-in>" 1
# 1 "<built-in>" 3
# 466 "<built-in>" 3
# 1 "<command line>" 1
# 1 "<built-in>" 2
# 1 "a.c" 2



"WEBP_MAX_DIMENSION"




"123"

This shows up in Pillow e.g. here (#4700):

Pillow/src/_imaging.c

Lines 4372 to 4374 in 4b03987

	#define tostr1(a) #a
	#define tostr(a) tostr1(a)
	PyObject *v = PyUnicode_FromString(tostr(LIBJPEG_TURBO_VERSION));

[radarhere]

Why does HandleEncoderError need to be a separate function?

[akx]

It doesn't, but it follows the form of HandleMuxError, above. Would you prefer it inlined?

[akx]

I'm reluctant to remove the test, since it is guarding against regression for the specific case of #5461.

AISI, the issue there is the WebP encoder not having reported even an error code? As noted in the PR description, with this PR that user will get "encoding error 6: partition #0 is bigger than 512K" (instead of "encoding error 6" as of #5471).

Since the remaining test covers the entire error handling code, IMO this PR covers that case too but without having to allocate a large image the WebP encoder can't fit in partition #0 (whatever that means 😅).

I'm having a hard time seeing the way this'd regress. Someone would need to explicitly change the error handling in the WebP encoder for worse?

EDIT: #9936 will enable the knob to tweak partition_limit, as noted in that issue 😄