Обсуждение: Python installation selection in Meson

I noticed that under meson, the selection of the Python installation 
using the 'PYTHON' option doesn't work completely.  The 'PYTHON' option 
determined the Python binary that will be used to call the various build 
support programs.  But it doesn't affect the Python installation used 
for PL/Python.  For that, we need to pass the program determined by the 
'PYTHON' option back into the find_installation() routine of the python 
module.  (Otherwise, find_installation() will just search for an 
installation on its own.)  See attached patch.  I ran this through 
Cirrus, seems to work.

Hi,

On 2023-11-28 19:02:42 +0100, Peter Eisentraut wrote:
> I noticed that under meson, the selection of the Python installation using
> the 'PYTHON' option doesn't work completely.  The 'PYTHON' option determined
> the Python binary that will be used to call the various build support
> programs.  But it doesn't affect the Python installation used for PL/Python.
> For that, we need to pass the program determined by the 'PYTHON' option back
> into the find_installation() routine of the python module.  (Otherwise,
> find_installation() will just search for an installation on its own.)  See
> attached patch.  I ran this through Cirrus, seems to work.

Makes sense!

Greetings,

Andres Freund

On 2023-11-28 Tu 13:02, Peter Eisentraut wrote:
> I noticed that under meson, the selection of the Python installation 
> using the 'PYTHON' option doesn't work completely.  The 'PYTHON' 
> option determined the Python binary that will be used to call the 
> various build support programs.  But it doesn't affect the Python 
> installation used for PL/Python.  For that, we need to pass the 
> program determined by the 'PYTHON' option back into the 
> find_installation() routine of the python module.  (Otherwise, 
> find_installation() will just search for an installation on its own.)  
> See attached patch.  I ran this through Cirrus, seems to work.


I noticed when working on the meson/windows stuff that meson would try 
to build plpython against its python installation, which failed 
miserably. The workaround was to abandon separate meson/ninja 
installations via chocolatey, and instead install them using pip. Maybe 
this was as a result of the above problem?


cheers


andrew

--
Andrew Dunstan
EDB: https://www.enterprisedb.com

On 29.11.23 14:23, Andrew Dunstan wrote:
> 
> On 2023-11-28 Tu 13:02, Peter Eisentraut wrote:
>> I noticed that under meson, the selection of the Python installation 
>> using the 'PYTHON' option doesn't work completely.  The 'PYTHON' 
>> option determined the Python binary that will be used to call the 
>> various build support programs.  But it doesn't affect the Python 
>> installation used for PL/Python.  For that, we need to pass the 
>> program determined by the 'PYTHON' option back into the 
>> find_installation() routine of the python module.  (Otherwise, 
>> find_installation() will just search for an installation on its own.) 
>> See attached patch.  I ran this through Cirrus, seems to work.
> 
> I noticed when working on the meson/windows stuff that meson would try 
> to build plpython against its python installation, which failed 
> miserably. The workaround was to abandon separate meson/ninja 
> installations via chocolatey, and instead install them using pip. Maybe 
> this was as a result of the above problem?

That sounds like it could be the case.

On 28.11.23 19:16, Andres Freund wrote:
> On 2023-11-28 19:02:42 +0100, Peter Eisentraut wrote:
>> I noticed that under meson, the selection of the Python installation using
>> the 'PYTHON' option doesn't work completely.  The 'PYTHON' option determined
>> the Python binary that will be used to call the various build support
>> programs.  But it doesn't affect the Python installation used for PL/Python.
>> For that, we need to pass the program determined by the 'PYTHON' option back
>> into the find_installation() routine of the python module.  (Otherwise,
>> find_installation() will just search for an installation on its own.)  See
>> attached patch.  I ran this through Cirrus, seems to work.
> 
> Makes sense!

I have committed this, and also backpatched to 16 to keep the behavior 
consistent.