py38-sip +universal: fatal error: 'stdio.h' file not found

[ryandesign (Ryan Carsten Schmidt)]

Building py38-sip with the +universal variant fails because it uses an invalid -isysroot:

:info:configure MacOS/X universal binaries will be created using
:info:configure /Applications/Xcode.app/Contents/Developer/Platforms/MacOSX.platform/Developer/SDKs/.

:info:build /usr/bin/clang -c -arch x86_64 -arch i386 -arch x86_64 -arch i386 -isysroot /Applications/Xcode.app/Contents/Developer/Platforms/MacOSX.platform/Developer/SDKs/ -Os -Wall -W -DNDEBUG -I. -o main.o main.c
:info:build main.c:20:10: fatal error: 'stdio.h' file not found
:info:build #include <stdio.h>
:info:build          ^~~~~~~~~
:info:build 1 error generated.

[kencu (Ken)]

we were just waiting for this to happen -- base has a nasty habit of leaving configure.sdkroot empty.

<​https://github.com/macports/macports-ports/commit/edcd25c14ead450a9e636bb136048f8bc2934b76#diff-5deb2be1b6e1dada51e71b2fde701626>

[ryandesign (Ryan Carsten Schmidt)]

It works fine when the universal variant is not used, so I assume py-sip is doing something weird, and that this is not related to what you're talking about in MacPorts base.

[kencu (Ken)]

I suspect it is:

--arch=i386 --sdk= LFLAGS="-F/opt/local/Library/Frameworks 

[kencu (Ken)]

I will try it with the fix and see if it works.

[ryandesign (Ryan Carsten Schmidt)]

My point is that --sdk= works fine for the non-universal build.

​https://build.macports.org/builders/ports-10.13_x86_64-builder/builds/89583/steps/install-port/logs/stdio

And the non-universal build does not say:

MacOS/X universal binaries will be created using

That's something specific to sip.

​https://www.riverbankcomputing.com/hg/sip/file/tip/configure.py.in

[kencu (Ken)]

base leaves configure.sdkroot empty when it should be "/". Most ports therefore test for this and set it to "/" if it's empty.

if --sdk is empty then sip goes down it's own pathways for finding sdks, some of which are different when building universal, and fails.

Too tired tonight; will build it tomorrow to confirm.

[kencu (Ken)]

add this:

    # work around base bug
    if {${configure.sdkroot} eq ""} {
        set configure.sdkroot "/"
    }

port builds:

$ port -v installed py38-sip
The following ports are currently installed:
  py38-sip @4.19.24_0+universal (active) platform='darwin 17' archs='i386 x86_64' date='2020-08-17T09:16:56-0700'

[kencu (Ken)]

<​https://github.com/macports/macports-ports/pull/8136>

Now -- if you also want to go down the rabbit hole of working with py-sip to fix their own version of their SDK-finding code, feel free. MacPorts has been using it's own for years, however, and even if they do fix theirs, we'll not use it (or at least IMHO we shouldn't start using it just to try to prove to me that we could).

[kencu (Ken)]

In 1d2ab119635c410e1415a6061e1dea5c6f4dee14/macports-ports (master):

py-sip: work around base bug

configure.sdkroot should never be ""
closes: #61004
see: ​https://github.com/macports/macports-base/pull/183

[michaelld (Michael Dickens)]

Merged; thx! I think this would be much better fixed in base (or some other common location, not individual Portfile's) ... and then we could remove these SDK tweaks from all of the Portfile's that do it ... there has got to be a better way!

[ryandesign (Ryan Carsten Schmidt)]

We've been using these workarounds in ports for awhile and I don't remember why. I agree it would be better to fix it centrally, but I assume there was a reason why we didn't do that (did it break something else?)