This is the mail archive of the gcc-patches@gcc.gnu.org mailing list for the GCC project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: [patch] Remove superfluous /dev/null on grep line


On 06/04/16 11:01 +0200, Jakub Jelinek wrote:
On Wed, Apr 06, 2016 at 09:50:48AM +0100, Jonathan Wakely wrote:
On 06/04/16 09:39 +0200, Eric Botcazou wrote:
>we recently ran into build failures on Windows systems using a somewhat old
>grep, coming from a syntax error in the libstdc++-symbols.ver version file:
>
># Symbol versioning for shared libraries.
>if ENABLE_SYMVERS
>libstdc++-symbols.ver:  ${glibcxx_srcdir}/$(SYMVER_FILE) \
>		$(port_specific_symbol_files)
>	cp ${glibcxx_srcdir}/$(SYMVER_FILE) $@.tmp
>	chmod +w $@.tmp
>	if test "x$(port_specific_symbol_files)" != x; then \
>	  if grep '^# Appended to version file.' \
>	       $(port_specific_symbol_files) /dev/null > /dev/null 2>&1; then
>\
>	    cat $(port_specific_symbol_files) >> $@.tmp; \
>	  else \
>	    sed -n '1,/DO NOT DELETE/p' $@.tmp > tmp.top; \
>	    sed -n '/DO NOT DELETE/,$$p' $@.tmp > tmp.bottom; \
>	    cat tmp.top $(port_specific_symbol_files) tmp.bottom > $@.tmp; \
>	    rm tmp.top tmp.bottom; \
>	  fi; \
>	fi
>
>Note the double /dev/null on the grep command line.  The first one causes the
>grep to fail when the command is invoked on these systems.  That's old code,
>but it is now invoked for config/abi/pre/float128.ver on the mainline and 5
>branch and this breaks the build on these systems (4.9 builds fine).
>
>This first /dev/null doesn't serve any useful purpose and seems to be a typo,

Doesn't it mean that if $port_specific_symbol_files contains only
whitespace we don't hang waiting for input from stdin? The 'if' above
it will be true when "x$port_specific_symbol_files" = "x " or similar.

I don't see any way for that to happen in the FSF tree, so it should
be safe. I'm a bit concerned about making that change this late in
stage 4 though. There isn't much time to find out if it breaks an
obscure target.

As it is a make variable, can't make be used to test this?
So perhaps
	chmod +w $@.tmp
ifneq ($(port_specific_symbol_files),)
	  if grep '^# Appended to version file.' \
	       $(port_specific_symbol_files) /dev/null > /dev/null 2>&1; then \
	    cat $(port_specific_symbol_files) >> $@.tmp; \
	  else \
	    sed -n '1,/DO NOT DELETE/p' $@.tmp > tmp.top; \
	    sed -n '/DO NOT DELETE/,$$p' $@.tmp > tmp.bottom; \
	    cat tmp.top $(port_specific_symbol_files) tmp.bottom > $@.tmp; \
	    rm tmp.top tmp.bottom; \
	  fi;
endif
?  Though, I think the initial and trailing whitespace is removed during
expansion (or already parsing of the vars), so even the
test "x$(port_specific_symbol_files)" != x
check should work right.

OK, I have no objection to the original patch then.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]