ext/uri: Fixes inconsistent validation ordering

[NickSdot]

Aligns builder and parser error order. Parser is LIFO, which is consistent with Lexbor logs (but imo surprising).

[NickSdot]

Does this need INTERNALS or something?

[TimWolla]

Does this need INTERNALS or something?

Not for static functions, since they are not part of the API / ABI.

[kocsismate]

@TimWolla I realized that the fact that the whatwg errors are LIFO is just a side effect of my unfortunate choice to use lexbor_array_obj_pop() back then to retrieve the logs from lexbor. It's possible to implement this retrieval in FIFO order. Something like:

ZEND_ATTRIBUTE_NONNULL static const char *fill_errors_inner(HashTable *errors)
{
	const char *result = NULL;
	lexbor_plog_t *log = lexbor_parser.log;
	size_t length = lexbor_plog_length(log);

	for (size_t i = 0; i < length; i++) {
		const lexbor_plog_entry_t *lxb_error = lexbor_array_obj_get(&log->list, i);

		const char *reason;
		if (append_validation_error(errors, lxb_error->id, (const char *) lxb_error->data, &reason) && result == NULL) {
			result = reason;
		}
	}

	lexbor_plog_clean(log);

	return result;
}

So the main question is if we are allowed to/want to change the order now or later? If yes, then as far as I can see, this PR should rather pivot to fix this detail.

[TimWolla]

So the main question is if we are allowed to/want to change the order now or later?

I don't think we ever guaranteed a specific order: Users are expected to check for ->failure to find out which error caused the parsing to abort. In practice I also suspect that this property is not really inspected in a programmatic fashion (versus just stashing it away in some log).

For PHP 8.6, I'd say: Ask the release managers (by requesting review from the team). For master / 8.7, this can just change with a UPGRADING note.

[NickSdot]

As mentioned in the desc, LIFO surprising. Should be FIFO; so keen to pivot. When there is a decision feel free to @ me.

[kocsismate]

@mbeccati @svpernova09 can you please chime in here? #23823 (comment) We would like to reverse the order of errors passed back/thrown by Uri\WhatWg\UrlBuilder, Uri\WhatWg\Url::__construct(), Uri\WhatWg\Url::parse(), as well as Uri\WhatWg\Url::set*(), since currently these errors are returned in backward order.

Can we still do it? As Tim wrote above, we guaranteed no particular order, and I agree with him that likely there is no much use of these errors besides displaying.

[mbeccati]

@kocsismate: we've discussed and we think it's ok to merge.

[kocsismate]

@NickSdot Can you please update this PR to fix the order of errors?

[NickSdot]

@NickSdot Can you please update this PR to fix the order of errors?

@kocsismate done and targeted 8.6

[kocsismate]

Overall: looks good, but I had a few nits before merge