Regular crash in modifying regex

[Richard Kimberly Heck]

On 1/14/21 2:15 PM, Richard Kimberly Heck wrote:
> On 1/14/21 1:24 PM, Kornel Benko wrote:
>> Am Thu, 14 Jan 2021 13:08:24 -0500
>> schrieb Richard Kimberly Heck <rikiheck at lyx.org>:
>>
>>> On 1/14/21 4:40 AM, Kornel Benko wrote:
>>>> Am Thu, 14 Jan 2021 04:29:21 -0500
>>>> schrieb Richard Kimberly Heck <rikiheck at lyx.org>:
>>>>  
>>>>> On 1/14/21 4:19 AM, Kornel Benko wrote:  
>>>>>> Am Thu, 14 Jan 2021 04:15:18 -0500
>>>>>> schrieb Richard Kimberly Heck <rikiheck at lyx.org>:
>>>>>>    
>>>>>>> On 1/14/21 3:45 AM, Kornel Benko wrote:    
>>>>>>>> Am Thu, 14 Jan 2021 07:53:16 +0100
>>>>>>>> schrieb Jürgen Spitzmüller <spitz at lyx.org>:
>>>>>>>>      
>>>>>>>>> Am Donnerstag, dem 14.01.2021 um 00:58 +0100 schrieb Kornel Benko:      
>>>>>>>>>> How to reproduce
>>>>>>>>>>
>>>>>>>>>> 1.) Open lyx file e.g. Additional.lyx
>>>>>>>>>> 2.) select AdvFind
>>>>>>>>>> 3.) select regexp-mode
>>>>>>>>>> 4.) insert any text not in this file
>>>>>>>>>> 5.) select search for section
>>>>>>>>>> 6.) change to not ignore format
>>>>>>>>>> 7.) start search
>>>>>>>>>> 8.) cancel the search (with escape)
>>>>>>>>>> 9.) clean the regex content and try to insert anything ==> crash
>>>>>>>>>>
>>>>>>>>>> BTW, lyx2.3 does not crash.        
>>>>>>>>> Does not crash for me. Can you produce a backtrace?
>>>>>>>>>
>>>>>>>>> Jürgen
>>>>>>>>>      
>>>>>>>> Yes, attached. Last commit was fcea6c47.      
>>>>>>> That is even weirder. You're getting an assertion, on the ground that a
>>>>>>> TOC is being requested that does not yet exist, and the type is
>>>>>>> "tableofcontents", which is supposed always to exist. But maybe that
>>>>>>> makes sense if it is the TOC for the F&R Buffer itself that is being
>>>>>>> queried: Probably it does not have a TOC at all.
>>>>>>>
>>>>>>> You might try reverting the problematic commit and see if that helps
>>>>>>> (i.e., if master minus that commit asserts). That would at least give us
>>>>>>> another data point.
>>>>>>>
>>>>>>> Oh, I see it! The TocBackend::toc method has two versions: A const one
>>>>>>> and a non-const one. Because of the const patch, the const one is being
>>>>>>> called, I think. It asserts when the requested TOC does not exist,
>>>>>>> whereas the other one just creates it and moves along. Though now I'm
>>>>>>> puzzled why I don't get a crash. But that makes me think that this has
>>>>>>> just revealed some other bug.    
>>>>>> Good news :)    
>>>>> OK, the assertion happens for you because this test
>>>>>
>>>>> master != buffer() && !master->hasGuiDelegate()
>>>>>
>>>>> passes for you and fails for me. I'm not sure why it would pass, though.
>>>>> I'm especially puzzled because the active Buffer at #13 in your
>>>>> backtrace is the same as what's reported as master at #8. So the test
>>>>> ought to fail.
>>>>>
>>>>> Riki
>>>>>
>>>>>  
>>>> I understood the backtrace so that the crash is _before_ this
>>>> 	master != buffer() && !master->hasGuiDelegate()
>>>> statement, namely in TocBackend::updateItem(). (This is one line above)  
>>> Oh, yes, you are right. Of course, that still leaves me puzzled why I
>>> can't reproduce.
>> Maybe you have not completely follow the recipe?
>> These are the essentials:
>>  Regex is wrapped into Chapter or Part environment
>>  format search is enabled
>>  the search is cancelled before found anything
>> Now any key which would insert into regex led to crash.
> OK, I didn't get the Chapter bit. Now I'm getting a much easier
> assertion: New file; open Adv F&R; insert section; type anything. Boom!

Should be fixed.

Riki