Re: Patchy email

[David Kastrup]

Knut Petersen <address@hidden> writes:

> On 29.07.19 17:55, David Kastrup wrote:
>> Well, it is rather obvious that you should not be seeing failure with
>> unchanged master (assuming that you have some reasonably current Python
>> on your system) and I should not be seeing failure with your patch
>> (assuming that it tests out on your system: the symptoms are just too
>> clear to seem dependent on Python version).  So it's likely that some
>> preinstalled part (assuming both of us worked from clean trees) creeps
>> into both our operations.  That would be more likely than not a
>> preexisting problem not caused by your patch but we would still better
>> fix it before it causes more headaches.
>
> In the chapter "Regtest comparison" of our contributor's guide we can read:
>
>    Run |make|with current git master without any of your changes.
>    Before making changes to the code, establish a baseline for the comparison 
> by going to the ‘$LILYPOND_GIT/build/’ directory and running:
>
>        make test-baseline
>
> Stracing shows that after 'make all' it is also necessary to execute
> 'make install' prior to 'make test-baseline' or 'make test-clean'.

That's insane.  Having to install LilyPond before being able to test it
makes no sense.

> If 'make install' is omitted, the result depends both on files from
> the baseline version and on files that belong to the version of
> lilypond that was last installed with the same prefix.

Then we need to fix that.

> David, please rerun the test of my patch against origin/master on your
> system. On my system the process succeeds now:

> If the process also succeeds on your system we would know that patchy
> needs a fix.

No, that make test needs a fix.  make test is supposed to test the
version in the work directory, not some weird amalgamate between system
installed version and work directory version.

Any idea what would be required here?

-- 
David Kastrup