Re: pg_regress cleans up tablespace twice.

On Sat, Jun 20, 2020 at 09:33:26AM +1200, Thomas Munro wrote:
> Hmm, somehow this (well I guess it's this commit based on timing and
> the area it touches, not sure exactly why) made cfbot's Windows build
> fail, like this:
> 
> --- C:/projects/postgresql/src/test/regress/expected/tablespace.out
> 2020-06-19 21:26:24.661817000 +0000
> +++ C:/projects/postgresql/src/test/regress/results/tablespace.out
> 2020-06-19 21:26:28.613257500 +0000
> @@ -2,83 +2,78 @@
> CREATE TABLESPACE regress_tblspacewith LOCATION
> 'C:/projects/postgresql/src/test/regress/testtablespace' WITH
> (some_nonexistent_parameter = true); -- fail
> ERROR: unrecognized parameter "some_nonexistent_parameter"
> CREATE TABLESPACE regress_tblspacewith LOCATION
> 'C:/projects/postgresql/src/test/regress/testtablespace' WITH
> (random_page_cost = 3.0); -- ok
> +ERROR: could not set permissions on directory
> "C:/projects/postgresql/src/test/regress/testtablespace": Permission
> denied
> 
> Any ideas?  Here's what it does:
> 
> https://github.com/macdice/cfbot/tree/master/appveyor

I am not sure, and I am not really familiar with this stuff.  Your
code does a simple vcregress check, and that should take care of
automatically cleaning up the testtablespace path.  The buildfarm uses
this code for MSVC builds and does not complain, nor do my own VMs
complain.  A difference in the processing after 2b2a070d is that the
tablespace cleanup/creation does not happen while holding a restricted 
token [1] anymore because it got out of pg_regress.c.  Are there any
kind of restrictions applied to the user running appveyor on Windows?

[1]: https://docs.microsoft.com/en-us/windows/win32/secauthz/restricted-tokens
--
Michael