Thread: master fails to build on Windows

On 5/18/18, Sandeep Thakkar <sandeep.thakkar@enterprisedb.com> wrote:
> Hi
>
> I was building the sources I got from https://www.postgresql.org/
> ftp/snapshot/dev/ on Windows x64 and got the whole bunch of errors
> like "*Cannot
> open include file: 'catalog/pg_type_d.h"* , *"Cannot open include file:
> 'catalog/pg_tablespace_d.h'" . *I've attached the log.

I'm thinking the include/catalog/pg_*_d.h headers got dist-cleaned
when creating the snapshot, and the build script doesn't know to copy
them over. If you're able to test, does the attached patch fix the
issue? (This may not be the right permanent fix, but hopefully it will
confirm my suspicion)

-John Naylor

John Naylor <jcnaylor@gmail.com> writes:
> On 5/18/18, Sandeep Thakkar <sandeep.thakkar@enterprisedb.com> wrote:
>> I was building the sources I got from https://www.postgresql.org/
>> ftp/snapshot/dev/ on Windows x64 and got the whole bunch of errors
>> like "*Cannot
>> open include file: 'catalog/pg_type_d.h"* , *"Cannot open include file:
>> 'catalog/pg_tablespace_d.h'" . *I've attached the log.

> I'm thinking the include/catalog/pg_*_d.h headers got dist-cleaned
> when creating the snapshot, and the build script doesn't know to copy
> them over. If you're able to test, does the attached patch fix the
> issue? (This may not be the right permanent fix, but hopefully it will
> confirm my suspicion)

Mmm, yeah, this is clearly an oversight.  I think we (or at least I)
figured it was okay to skip the separate stamp file and test
for the install-the-headers step because the Windows scripts didn't
have a way to blow away the include/ headers without deleting the
underlying backend/ headers --- but looking closer at clean.bat,
that's quite wrong.  clean.bat will do it by default :-(.  So I think
the snapshot was probably fine, but Sandeep ran clean.bat and then
things weren't fine.

It seems like clean.bat's default behavior doesn't correspond very well
to any of the three cleaning levels that the Makefiles understand,
while its "dist" behavior seems to be "make maintainer-clean", not
"make distclean" as one might expect.  Maybe that's something to
improve someday, but right now is probably not the time.

In the meantime, I think your patch is a good idea whether or not
this is the right explanation for Sandeep's problem.  The closer
the Windows and Unix build rules are, the better.

            regards, tom lane