Re: Compile Error v2.13.2 on Solaris SPARC

"Liam R. Howlett" [off-list ref] writes:

I'd say it is a bad idea to define a symbol that you _know_ a
platform header file defines.  Any header you may include from the
platform after these lines still expects the symbol to be defined in
a way it defines without getting molested and will misbehave.

* Junio C Hamano [off-list ref] [170627 13:38]:

"Liam R. Howlett" [off-list ref] writes:

quoted

How about:

---- 8< ----

diff --git a/sha1dc/sha1.c b/sha1dc/sha1.c
index facea1bb5..ed8c63f2d 100644
--- a/sha1dc/sha1.c
+++ b/sha1dc/sha1.c

@@ -38,9 +38,18 @@
 
 #if (defined(_BYTE_ORDER) || defined(__BYTE_ORDER) || defined(__BYTE_ORDER__))
 
+#define EMPTY_VAL(x) x ## 1
+#define EMPTY(x) EMPTY_VAL(x)
+
+#if (defined(_BIG_ENDIAN) && (EMPTY(_BIG_ENDIAN) == 1))
+#undef _BIG_ENDIAN
+#define _BIG_ENDIAN 4321
+#endif

I'd say it is a bad idea to define a symbol that you _know_ a
platform header file defines.  Any header you may include from the
platform after these lines still expects the symbol to be defined in
a way it defines without getting molested and will misbehave.

Okay.  Thanks.  I thought a c file would be safe, especially after the
includes but there is indeed a possible include (ifdef'ed) later.

This compressed logic is causing a lot of issues.  Could we just rewrite
it as a whole lot of #if/#else, statements to avoid running across the
issue where the precompiler does not short-circuit the checks?  Would
this cause any other issues?

Alternatively, we can replace the undef/define with the define of
SHA1DC_BIGENDIAN and make this an #if/#else..

A third option is to compile a small test and just -DSHA1DC_BIGENDIAN in
the Makefile.

Cheers,
Liam

On Tue, Jun 27 2017, Liam R. Howlett jotted:

This compressed logic is causing a lot of issues.  Could we just rewrite
it as a whole lot of #if/#else, statements to avoid running across the
issue where the precompiler does not short-circuit the checks?  Would
this cause any other issues?

Again, this is hopefully addressed by my
20170627121718.12078-2-avarab@gmail.com ([PATCH 1/3] sha1dc: update from
my PR #36).

A third option is to compile a small test and just -DSHA1DC_BIGENDIAN in
the Makefile.

This would be ideal, but so far the only facility we have for that is
the configure script, which there have been objections to making a hard
dep in the past, thus we have various bits done via macros that would be
better done via built-time compiling & testing a C program.

My memory of such discussions is hazy though, did people fundimentally
object to the idea, or was it just an objection to autoconf in
particular, I don't know.

If it was just autoconf maybe someone more clever at Makefile magic than
me could come up with a way to compile a test program that would then
define a flag that would be passed to the rest of the programs, and set
up dependencies in such a way that it was done before anything else, I
don't know if that's easy/possible with make.