Promote warnings to exceptions in ext/intl

[kocsismate]

There is quite a few intl_error_set() invocations remaining, out of which probably a few could also be promoted to an exception.

[kocsismate]

Hmm, there is no such warning in strpos(), so maybe we could get rid of this error? On the other hand, the offset is not contained... case has been promoted to an exception in strpos(), so I'll do it here as well.

[nikic]

Yes, we should add support for empty string here.

[nikic]

Yes, we should add support for empty string here.

[nikic]

Marking as reviewed.

[nikic]

Shouldn't this one also be string(3) "abc"?

[kocsismate]

Ah, yes.. I thought that the removed check from strstr_common_handler() was inside if ( !f_ignore_case ) {}, thus it only applies to grapheme_strstr. And when I double-checked it on 3v4n, the intl.error_level tricked me, since there is no error shown by default.

That said, I'll try to fix this.

[kocsismate]

The reason of the false return value is this check in grapheme_strpos_utf16():

src = usearch_open(uneedle, uneedle_len, uhaystack, uhaystack_len, "", bi, &status);
STRPOS_CHECK_STATUS(status, "Error creating search object");

To make things worse, grapheme_strpos_utf16() is called in the other grapheme string functions when the offset is below 0, so e.g. grapheme_strpos("abc", "", -1) also fails currently.

I don't see any easy solution for the problem because usearch_open() is third-party code, otherwise we could just add a similar check for empty needle what George added to zend_memnstr(). @nikic Do you know about any good solution that circumvents this problem? Is it a very big problem if grapheme_() functions can't handle empty needles?

[nikic]

Looks like the relevant icu code is: https://github.com/unicode-org/icu/blob/f744742e036441cbfd54b535e722dc3801ff0f63/icu4c/source/i18n/usearch.cpp#L2675

I think we should add an explicit check for this case then. I do think it would be good to support empty strings here, as we also added support for them in ext/standard and ext/mbstring (I'm not sure about ext/iconv right now).

[kocsismate]

I've reverted the change in question, but added a task

[nikic]

Travis still failing.

[kocsismate]

@nikic Yeah, I'll fix it too. It didn't run on my machine due to incompatible icu version.