msvc: fix -I"" for same directory pch header

[Kojoley]

It would be nice to fix that by doing :D:E=. but unfortunately it does not work.

Fixes #229

[Kojoley]

Blah, this will blow up if it will expand to an absolute path, any ideas of a simple fix? Workarounding the lack of :D:E=. requires a lot of boilerplate here :-(

[Kojoley]

I would say that the bug is in B2 :D expanding, filed #232.

[grafikrobot]

The only semi non-hack solution I can think of is doing the dir/file split in the rule instead of the action. That way the dir will end up as an empty list instead of the empty string in the list:

rule compile-c-c++ ( targets + : sources * )
{
    DEPENDS $(<[1]) : [ on $(<[1]) return $(PCH_HEADER) ] ;
    DEPENDS $(<[1]) : [ on $(<[1]) return $(PCH_FILE) ] ;
    PDB_NAME on $(<) = $(<[1]:S=.pdb) ;
    LOCATE on $(<[1]:S=.pdb) = [ on $(<[1]) return $(LOCATE) ] ;
    PCH_HEADER_DIR on $(<) = $(PCH_HEADER:D) ;
    PCH_HEADER_DIR on $(<) ?= "." ;
    PCH_HEADER on $(<) = $(PCH_HEADER:D=) ;
}

And the action uses -I"$(PCH_HEADER_DIR)" -FI"$(PCH_HEADER)".

[Kojoley]

That what I called a workaround :-) I would prefer :D or :D:E=. to just work...

[Kojoley]

The only semi non-hack solution I can think of is doing the dir/file split in the rule instead of the action. That way the dir will end up as an empty list instead of the empty string in the list:

rule compile-c-c++ ( targets + : sources * )
{
    DEPENDS $(<[1]) : [ on $(<[1]) return $(PCH_HEADER) ] ;
    DEPENDS $(<[1]) : [ on $(<[1]) return $(PCH_FILE) ] ;
    PDB_NAME on $(<) = $(<[1]:S=.pdb) ;
    LOCATE on $(<[1]:S=.pdb) = [ on $(<[1]) return $(LOCATE) ] ;
    PCH_HEADER_DIR on $(<) = $(PCH_HEADER:D) ;
    PCH_HEADER_DIR on $(<) ?= "." ;
    PCH_HEADER on $(<) = $(PCH_HEADER:D=) ;
}

And the action uses -I"$(PCH_HEADER_DIR)" -FI"$(PCH_HEADER)".

That won't work for two reasons:

• Empty variable i.e "" is not the same thing as unset, so ?= "." will do nothing here.
• Even if it worked for empties, there could be multiple precompile headers at the same time.

[Kojoley]

I realized it is might be wrong to create PCH with only header name, it is preferable to use the exact path the user typed, but it seems to hard to obtain it? Also made some refactoring as a drive by.

[Kojoley]

I'm actually not sure why it was done like this, maybe I miss something, like it's need to run header scanning or something else?

[grafikrobot]

Maybe header scanning. But not sure on that. It does seem rather weird.

[grafikrobot]

/backport

[github-actions]

Successfully created backport PR for version/4.10.0:

• [Backport version/4.10.0] msvc: fix -I"" for same directory pch header #247