Interdependent, failing commits

[Stephan Bergmann]

On 08.01.25 08:58, Stephan Bergmann wrote:
> On 1/6/25 22:27, Stephan Bergmann wrote:
>> The two recent commits <https://git.libreoffice.org/core/+/ 
>> d8e29e2697e90da6acf5858fd74d6054ed9e5aa3%5E%21> "Revert 'Revert 
>> 'stronger typing for SwClient::GetRegisteredIn'' and fix SwIterator 
>> cast" and <https://git.libreoffice.org/core/ 
>> +/404fbec25740406e3b926844f3bd0f324dc38b8c%5E%21> "convert 
>> RES_ATTRSET_CHG to SfxHint" are both known-broken and cause at least 
>> UBSan builds to fail (see the comments at <https:// 
>> gerrit.libreoffice.org/c/core/ 
>> +/179480/3#message-1d186afad4accdcf814507953175bf0e483b0e2e> and 
>> <https://gerrit.libreoffice.org/c/core/+/179772/6#message- 
>> e7324ab2f8feb8f37c28a4f8845176d82f658648>, respectively).
>>
>> Now, a third commit <https://git.libreoffice.org/core/ 
>> +/404fbec25740406e3b926844f3bd0f324dc38b8c%5E%21> "convert 
>> RES_ATTRSET_CHG to SfxHint" interferes with (local) reverts of the 
>> other two commits, so it becomes increasingly messy to try and keep a 
>> successful (local) UBsan build, and keep an eye on further commits 
>> that could potentially introduce further/unrelated breakage.
> 
> The above paragraph erroneously references the wrong commit.  The third 
> commit, which interferes with reverts of the other two commits, rather 
> is <https://gerrit.libreoffice.org/c/core/+/179781> "convert 
> RES_OBJECTDYING to SfxHint".

And that third commit also turned out to cause further UBSan failures of 
its own, see the comment at 
<https://gerrit.libreoffice.org/c/core/+/179781/6#message-65721cd380de7681cd9ade644da755e6549e23f7> 
"convert RES_OBJECTDYING to SfxHint"-

>> Therefore, unless there are fixes for the two known-broken commits 
>> available by then, I suggest to revert for now all three of those 
>> commits tomorrow evening.
> 
> (Before I could exercise that plan, things started to get even more 
> messy, with more new and interdependent commits getting thrown into the 
> fire, causing further UBSan failures, and some of the UBSan failures 
> seemingly only experienced with some builds but not with others.  Stay 
> tuned...)

So I'm now at a chain of four reverts
* <https://gerrit.libreoffice.org/c/core/+/179940/> "Revert 'fix ubsan 
failure'"
* <https://gerrit.libreoffice.org/c/core/+/179925/> "Revert 'convert 
RES_OBJECTDYING to SfxHint'"
* <https://gerrit.libreoffice.org/c/core/+/179941/> "Revert 'convert 
RES_ATTRSET_CHG to SfxHint'"
* <https://gerrit.libreoffice.org/c/core/+/179946/> "Revert 'Revert 
'Revert 'stronger typing for SwClient::GetRegisteredIn'' and fix 
SwIterator cast'"
that will together hopefully turn 
<https://ci.libreoffice.org/job/lo_ubsan/> green again.  (I know such a 
mass revert orgy is not exactly ideal, but it appears to me the only 
viable way forward at the moment.)

Björn, Noel, I would be grateful if you could refrain from pushing 
further commits in those areas until the above Gerrit chain has been 
pushed and a fresh build of <https://ci.libreoffice.org/job/lo_ubsan/> 
(which I'll trigger once the chain has been pushed) is seen green 
(hopefully).  Thanks