SIGSEGV if `Match` is called before `Compile`

[masklinn]

>>> f = Filter()
>>> f.Compile()
re2/filtered_re2.cc:74: Compile called before Add.
>>> f.Match("")

>>> f = Filter()
>>> f.Match("")
py311: exit -11

This is inside tox, using python installed with pyenv, but reproduces with both Python 3.8 and 3.11.

I would assume it's because while there is a guard to check that the filter is compiled in FirstMatch

re2/re2/filtered_re2.cc

Lines 96 to 99 in 108914d

	if (!compiled_) {
	LOG(DFATAL) << "FirstMatch called before Compile.";
	return -1;
	}

there is no such guard for AllMatches:

re2/re2/filtered_re2.cc

Lines 108 to 118 in 108914d

	bool FilteredRE2::AllMatches(absl::string_view text,
	const std::vector<int>& atoms,
	std::vector<int>* matching_regexps) const {
	matching_regexps->clear();
	std::vector<int> regexps;
	prefilter_tree_->RegexpsGivenStrings(atoms, &regexps);
	for (size_t i = 0; i < regexps.size(); i++)
	if (RE2::PartialMatch(text, *re2_vec_[regexps[i]]))
	matching_regexps->push_back(regexps[i]);
	return !matching_regexps->empty();
	}

I assume the PrefilterTree would be the issue but apparently it guards against this

re2/re2/prefilter_tree.cc

Lines 271 to 276 in 108914d

	if (!compiled_) {
	// Some legacy users of PrefilterTree call Compile() before
	// adding any regexps and expect Compile() to have no effect.
	// This kludge is a counterpart to that kludge.
	if (prefilter_vec_.empty())
	return;

so I'm not sure why it SIGSEGVs, but I have not debugged the C++ code just looked at it.

Interestingly there is a guard in Set::Match:

re2/re2/set.cc

Lines 128 to 133 in 108914d

	if (!compiled_) {
	if (error_info != NULL)
	error_info->kind = kNotCompiled;
	LOG(DFATAL) << "RE2::Set::Match() called before compiling";
	return false;
	}

but Filter::Match does not check it:

re2/python/_re2.cc

Line 226 in 108914d

set_->Match(text, &atoms);

so it does not protect the FilteredRE2 (assuming that's what crashes, for all I know it might be something else entirely).

[junyer]

Thanks for the report! This crash occurs here: set_ will be null until Compile() has been called. I think raising ValueError would be reasonable.

[junyer]

Having meditated on it a little more, raising re2.error would be preferable – and arguably trivial as per this documentation.

[masklinn]

Thanks for the report! This crash occurs here: set_ will be null until Compile() has been called.

Oh yeah that's obvious now that you point it out, I had not noticed the Set is only instantiated during compilation so I assumed that bit was fine and went on to look in the prefilter, sorry about that.

[junyer]

FYI, I just cut release 2024-04-01 (and published wheels) containing the fixes for this issue and also for issue #485. :)

[masklinn]

FYI, I just cut release 2024-04-01 (and published wheels) containing the fixes for this issue and also for issue #485. :)

Awesome, thanks.