Link error when compiling with -fsanitize=undefined

[Kornel Benko]

Am Mon, 20 Sep 2021 11:48:37 +0200
schrieb Kornel Benko <kornel at lyx.org>:

> Am Sun, 19 Sep 2021 20:19:15 -0400
> schrieb Scott Kostyshak <skostysh at lyx.org>:
> 
> > On Sun, Sep 19, 2021 at 08:37:16PM +0200, Kornel Benko wrote:  
> > > Am Sun, 19 Sep 2021 14:15:21 -0400
> > > schrieb Scott Kostyshak <skostysh at lyx.org>:
> > >     
> > > > On Sun, Sep 19, 2021 at 02:00:33PM +0200, Kornel Benko wrote:    
> > > > > Am Sun, 19 Sep 2021 07:45:19 -0400
> > > > > schrieb Scott Kostyshak <skostysh at lyx.org>:
> > > > >       
> > > > > > > > > So inserting this include cured the compilation for me.          
> > > > > > > > 
> > > > > > > > That's great you figured it out and it doesn't add too much complexity to
> > > > > > > > the code. Are you sure it ended up using the -fsanitize=undefined flag ?
> > > > > > > > If you do
> > > > > > > > 
> > > > > > > >   ldd build-dir/bin/lyx | grep ubsan
> > > > > > > > 
> > > > > > > > Does it show up?
> > > > > > > > 
> > > > > > > > Scott        
> > > > > > > 
> > > > > > > 	libubsan.so.1 => /usr/lib/x86_64-linux-gnu/libubsan.so.1
> > > > > > > (0x00007f49cdfd2000)        
> > > > > > 
> > > > > > Great, I will experiment next week with it.
> > > > > > 
> > > > > > Thank you,
> > > > > > 
> > > > > > Scott
> > > > > >       
> > > > > 
> > > > > BTW, how got you the output showing errors? My lyx does not print any error
> > > > > messages. (I mean:
> > > > > 'ColorCode' /home/scott/lyxbuilds/master/repo/src/MetricsInfo.cpp:174:6: runtime
> > > > > error: load of value 128, which is not a valid value for type
> > > > > ...
> > > > > )      
> > > > 
> > > > It seems that it is triggered with a lot of branches (or perhaps more generally
> > > > any different kind of insets?). Attached is an LFUN sequence that reproduces it.
> > > > Does it trigger the error for you?
> > > > 
> > > > Scott    
> > > 
> > > Not on empty document.    
> > 
> > For me it works on an empty document. I do use a default.lyx document (this is what
> > saves when you click on "save class defaults"), but I can reproduce if I remove that
> > template. I attach both MWEs; there's not much difference between them. For me, it is
> > sufficient to just open them and I get the messages in the terminal.
> > 
> > Not sure if it's relevant, but perhaps it could be differences with GCC version:
> > 
> >   g++ --version
> >   g++ (Ubuntu 10.3.0-1ubuntu1) 10.3.0
> > 
> > Scott  
> 
> My fail (Using LYX_DEBUG_SANITIZE too late). I get the messages now too.
> 
> BTW, I wonder, if if LYX_ASAN and LYX_DEBUG_SANITIZE are eventually exclusive?
> 
> 	Kornel

Some debugging lead to following observation:
Each inserted branch add a color to the list. The original enum defined colors (about
102), last Color_ignore.
So each branch will now get colors 103, 104, ..., and so on, so that the original enum
(probably used as 'char') cannot contain the value beyond 127).

Expanding the colors in src/ColorCode.h as in attached, removes the sanitize messages.

	Kornel
-------------- next part --------------
A non-text attachment was scrubbed...
Name: ColorCode.h.patch
Type: text/x-patch
Size: 389 bytes
Desc: not available
URL: <http://lists.lyx.org/pipermail/lyx-devel/attachments/20210920/2450d81c/attachment.bin>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 488 bytes
Desc: Digitale Signatur von OpenPGP
URL: <http://lists.lyx.org/pipermail/lyx-devel/attachments/20210920/2450d81c/attachment.asc>