ESC meeting minutes: 2020-09-10

Jan-Marek Glogowski glogow at fbihome.de
Mon Sep 14 11:55:10 UTC 2020



Am 14.09.20 um 13:39 schrieb Miklos Vajna:
> Hi Jan-Marek,
> 
> On Mon, Sep 14, 2020 at 12:48:54PM +0200, Jan-Marek Glogowski <glogow at fbihome.de> wrote:
>> I did Android builds, but I didn't change the config. If I change
>> autogen.input, I hit the same error.
> 
> OK, so it's not my broken setup, good.
> 
>> bibisecting points to commit 42aeb9f906ca4e23d118ff8563184f9315ef3b82. I
>> expected fallout from it, but this result is kind of a mystery. It seems
>> a revert of it fixes the problem.
>>
>> My current guess is, that my OVERWRITING_SETTINGS "shortcut" to prevent
>> going via *_FOR_BUILD variables somehow triggers this, but no idea how
>> and why... maybe some clashes with some internal configure state?
> 
> I guess there are 2 problems here:
> 
> 1) auto-re-running configure fails

That is a safety measure to prevent re-configuration in the different
makefile targets for the build and the host config. The whole cross
setup is done by a single configure call, which will run a sub-configure
for the build host to setup config_build.mk for the cross-tools. Your
"error" is correct to prevent an additional configure run.

> 2) even if you re-run configure manually, the auto-re-run will still
> happen.

That is ok too, because you can just run configure for the host system,
not the build system. What the "error" prevents is an automatic AKA make
triggered, 2nd re-run for the build system, which should never be needed.

> Not sure of both are a problem since the above commit. :-)
> 
> Perhaps 2) is easier to debug. I don't understand why we try to
> auto-re-run when autogen.sh was invoked manually. Does it not update
> the timestamp of some stamp file?

I have no idea yet, why this happens. All I currently know is, that
appealingly a revert of the patch fixes this and the only general change
I introduced in it, was the OVERWRITING_SETTINGS stuff, which interferes
between the host and build configuration runs. Nothing else in this
change looks suspicious. I still might be totally wrong, as I don't
know, what this might trigger to break in this way, or why.

Jan-Marek


More information about the LibreOffice mailing list