On 07/05/2010 09:55 AM, Diego E. "Flameeyes" Pettenò wrote:
> Il giorno lun, 05/07/2010 alle 16.44 +0000, Monty Taylor ha scritto:
>>
>> Yeah - so I see what you're doing, but I'm going to need to poke at
>> this some on the other platforms. I _really_ don't like trusting
>> things like pg_config.
>
> Not sure on Ubuntu but Gentoo tries to keep these scripts valid because
> they _are_ the only way upstream has to signal them..
Sure - the problem is that using autoconf at all is about making sure
that I can also build on broken platforms like OSX and Solaris... so
trusting that a script or a system config has been properly maintained
is just asking for failure. (This is the reason I hate pkg-config,
too... did they learn nothing from imake?)
It works _great_ when it works, but I _still_ need to test that it's not
lying or broken, which is why I'd prefer to just actually test the files
in the first place.
I'll see if I can merge the approaches though...
>> Also- why did you add and include of the pg_config.h include? That's
>> going to have values from their run of configure which we do not care
>> about?
>
> That's how most other projects get the correct build parameters, without
> this it fails here ... I sincerely forgot the error message right now.
I really need to blog more about this. It's so fundamentally wrong to do
that. Every project that does it is completely breaking everything.
MySQL does it - it was the first thing I fixed in Drizzle.
I guess it's time to go make a patch to postgres. :)
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.10 (GNU/Linux)
Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org/
-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1
On 07/05/2010 09:55 AM, Diego E. "Flameeyes" Pettenò wrote:
> Il giorno lun, 05/07/2010 alle 16.44 +0000, Monty Taylor ha scritto:
>>
>> Yeah - so I see what you're doing, but I'm going to need to poke at
>> this some on the other platforms. I _really_ don't like trusting
>> things like pg_config.
>
> Not sure on Ubuntu but Gentoo tries to keep these scripts valid because
> they _are_ the only way upstream has to signal them..
Sure - the problem is that using autoconf at all is about making sure
that I can also build on broken platforms like OSX and Solaris... so
trusting that a script or a system config has been properly maintained
is just asking for failure. (This is the reason I hate pkg-config,
too... did they learn nothing from imake?)
It works _great_ when it works, but I _still_ need to test that it's not
lying or broken, which is why I'd prefer to just actually test the files
in the first place.
I'll see if I can merge the approaches though...
>> Also- why did you add and include of the pg_config.h include? That's
>> going to have values from their run of configure which we do not care
>> about?
>
> That's how most other projects get the correct build parameters, without
> this it fails here ... I sincerely forgot the error message right now.
I really need to blog more about this. It's so fundamentally wrong to do
that. Every project that does it is completely breaking everything.
MySQL does it - it was the first thing I fixed in Drizzle.
I guess it's time to go make a patch to postgres. :) enigmail. mozdev. org/
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.10 (GNU/Linux)
Comment: Using GnuPG with Mozilla - http://
iEYEARECAAYFAkw yECAACgkQ2Jv7/ VK1RgErngCgvcFu XpV9ymns8Ya4fPp hgTY7 6Vg+ak2VbhRZxYF ZM
uLsAn3BRXSCCwiu
=YhS+
-----END PGP SIGNATURE-----