On 11/16/2015 05:53 PM, Guillaume Munch wrote:
> Le 16/11/2015 20:22, Georg Baum a écrit :
>> Richard Heck wrote:
>>
>>>
>>> Recipe:
>>>
>>> 1. Open the User Guide.
>>> 2. Click somewhere in the main text.
>>> 3. Insert > Citation
>>> 4. Enter the search field and type "cap"
>>> 5. Hit search button.
>>>
>>> Error: Software exception Detected
>>> ----------------------------------------
>>> LyX has caught an exception, it will now attempt to save all unsaved
>>> documents and exit.
>>>
>>> Exception: regex_error
>>>
>>> This is on Fedora 22, gcc 5.1.1, with configuration as:
>>> --enable-build-type=dev.
>>>
>>> The offending code is this assignment
>>>
>>> static const lyx::regex reg("[].|*?+(){}^$\\[\\\\]");
>>>
>>> in escape_special_characters in GuiCitation.cpp. I would guess that
>>> this
>>> is a consequence of changes to the regex engine.
>>
>> I can reproduce with gcc 5.1 in C++11-mode. This is most likely the same
>> problem as http://www.lyx.org/trac/ticket/9799, since the test fails
>> as well
>> for the same compiler, so I guess it happens with gcc4 in C++11 mode as
>> well. Nice to see that the unit test worked! Now we only need to
>> learn not
>> to ignore failing unit tests.
>>
>> Some regex expert needs to find out whether the regex syntax we are
>> using is
>> only valid for boost::regex or for C++11 std::regex as well, and
>> depending
>> on the outcome we either need to change our regex, or file a gcc bug. Or
>> maybe it is some multithreading issue? The 'static' rings some alarm
>> bells...
>>
>>
>
> If I may try to explain with my own words (because I really didn't get
> your reasoning at first, being far from being a C++ regex expert).
> Std::regex and boost::regex do not set the same regex style by default
> (perl vs ECMAScript) but both support other styles by passing the
> appropriate flag. The commit 4620034e already had fixed a crash a
> while ago due the use of (what is now) std::regex in MSVC, by passing
> automatically the flag to set the "grep" regex style instead of
> ECMAScript.
>
> But, in 394e1bf9 you did without passing this flag. The reason, if I
> reconstruct correctly, is that std's ECMAScript style is only supposed
> to be boost's perl style minus a few perl-isms; we shouldn't have to
> set a style "grep" (that may even change the meaning of regexes!). The
> most sensible cross-platform solution is to conform to the ECMAScript
> subset, and if I infer correctly this is what you mean.
>
> Once this is understood, we go to the ECMAScript standard and realize
> that "[]" is interpreted as the empty class in ECMAScript. "[]…]" is a
> perl-ism, and ] should be escaped.
>
> Lesson for everybody apart from Georg: please write regexes according
> to the ECMAScript standard in the future, even if boost is happy with
> your perl-isms.
>
> Here's a patch that locates a similar issue elsewhere. I did my fair
> share of the exercise, can I please let other people test it and
> commit the appropriate solution?
I can confirm that this fixes the original crash. I didn't check the
ExternalTransforms part of it, as I'm not sure how to do that, but one
would suppose it was right, too.
Richard
