Re: [RFC PATCH] Move SHA-1 implementation selection into a header file

["brian m. carlson" <sandals@crustytoothpaste.net>]

[-- Attachment #1: Type: text/plain, Size: 5169 bytes --]

On Sun, Mar 12, 2017 at 09:01:49AM -0400, Jeff King wrote:
> On Sat, Mar 11, 2017 at 10:28:18PM +0000, brian m. carlson wrote:
> 
> > Many developers use functionality in their editors that allows for quick
> > syntax checks, including warning about questionable constructs.  This
> > functionality allows rapid development with fewer errors.  However, such
> > functionality generally does not allow the specification of
> > project-specific defines or command-line options.
> > 
> > Since the SHA1_HEADER include is not defined in such a case, developers
> > see spurious errors when using these tools.  Furthermore, while using a
> > macro as the argument to #include is permitted by C11, it isn't
> > permitted by C89 and C99, and there are known implementations which
> > reject it.
> > 
> > Instead of using SHA1_HEADER, create a hash.h header and use #if
> > and #elif to select the desired header.  Have the Makefile pass an
> > appropriate option to help the header select the right implementation to
> > use.
> 
> This has bit me once or twice in the past[1], too, so I'm happy to see
> somebody tackling it.
> 
> Your solution does mean that your tool, whatever it is, that runs
> without the same CFLAGS as the Makefile may get a _different_ sha1
> implementation, though. That's probably good enough for some purposes
> but perhaps not for others.

Yeah, my goal was basically to pass -fsyntax-only, not to produce useful
object files.  My patch does basically require that the user have
OpenSSL installed, but I do, so it doesn't matter.

I considered after the fact that I might just do something like:

  #ifdef SHA1_HEADER
  #include SHA1_HEADER
  #else
  #include "block-sha1/sha1.h"
  #endif

That would be the smallest change, but probably not the best.

> The "most correct" solution, I think, would be to stop using "-D" flags
> in favor of actually generating hash.h. Something like the patch
> below[2] (though I think it there are some rough edges with the
> dependencies).

I agree that's a better solution.  I was concerned about avoiding ending
up rebuilding everything when we regenerated the file, but it looks like
you've avoided that with cmp.

> Of course the sha1 header is just one of many such defines. It's the one
> that is most visible because the result is syntactically valid without
> it, but anything you compile without the Makefile's CFLAGS may be subtly
> different than what the Makefile would produce. So arguably the Makefile
> should be writing out a build-options.h with all of the values, and that
> should get pulled in by git-compat-util.h.
> 
> I don't know if we want to go down that rabbit hole or not. I offer it
> merely as an alternative. I'm OK with your patch as-is if you don't want
> to dump any more time into it.

I'll take this patch for now and fix it up with the comment I mentioned
below.  If someone wants to improve the situation down the line, then
they can pick that up.

I assume I can apply your sign-off to the resulting patch?

> -Peff
> 
> [1] I think I hit the problem when trying to debug some internal part of
>     git and writing a one-off "foo.c" that calls the code I want. You
>     can't compile it with "gcc foo.c".
> 
> [2] Here's what a patch for the generated-header might look like:
> 
> diff --git a/Makefile b/Makefile
> index 9f0eae428..0d65d50e9 100644
> --- a/Makefile
> +++ b/Makefile
> @@ -690,6 +690,7 @@ XDIFF_LIB = xdiff/lib.a
>  VCSSVN_LIB = vcs-svn/lib.a
>  
>  GENERATED_H += common-cmds.h
> +GENERATED_H += hash.h
>  
>  LIB_H = $(shell $(FIND) . \
>  	-name .git -prune -o \
> @@ -1639,8 +1640,7 @@ PERLLIB_EXTRA_SQ = $(subst ','\'',$(PERLLIB_EXTRA))
>  # from the dependency list, that would make each entry appear twice.
>  LIBS = $(filter-out %.o, $(GITLIBS)) $(EXTLIBS)
>  
> -BASIC_CFLAGS += -DSHA1_HEADER='$(SHA1_HEADER_SQ)' \
> -	$(COMPAT_CFLAGS)
> +BASIC_CFLAGS += $(COMPAT_CFLAGS)
>  LIB_OBJS += $(COMPAT_OBJS)
>  
>  # Quote for C
> @@ -1781,7 +1781,7 @@ git$X: git.o GIT-LDFLAGS $(BUILTIN_OBJS) $(GITLIBS)
>  	$(QUIET_LINK)$(CC) $(ALL_CFLAGS) -o $@ $(ALL_LDFLAGS) \
>  		$(filter %.o,$^) $(LIBS)
>  
> -help.sp help.s help.o: common-cmds.h
> +help.sp help.s help.o: common-cmds.h hash.h
>  
>  builtin/help.sp builtin/help.s builtin/help.o: common-cmds.h GIT-PREFIX
>  builtin/help.sp builtin/help.s builtin/help.o: EXTRA_CPPFLAGS = \
> @@ -1805,6 +1805,10 @@ common-cmds.h: generate-cmdlist.sh command-list.txt
>  common-cmds.h: $(wildcard Documentation/git-*.txt)
>  	$(QUIET_GEN)$(SHELL_PATH) ./generate-cmdlist.sh command-list.txt >$@+ && mv $@+ $@
>  
> +hash.h:
> +	$(QUIET_GEN)echo '#include $(SHA1_HEADER)' >$@+ && \
> +	{ cmp $@+ $@ >/dev/null 2>/dev/null || mv $@+ $@; }

I think here we'd want to also "rm -f $@+", so that we don't leave an
extra file behind if we're up-to-date (which is the common case), much
like we do for GIT-BUILD-OPTIONS.
-- 
brian m. carlson / brian with sandals: Houston, Texas, US
+1 832 623 2791 | https://www.crustytoothpaste.net/~bmc | My opinion only
OpenPGP: https://keybase.io/bk2204

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 868 bytes --]