I have a GNU makefile. The makefile has a few tests (around line 65 from the linked file):

UNAME = uname

MACHINE ?= $(shell $(UNAME) -m)
SYSTEM ?= $(shell $(UNAME) -s)

IS_X86 = $(shell echo $MACHINE | $(EGREP) -c "i.86|x86|i86|amd64")
IS_X86_64 = $(shell echo $MACHINE | $(EGREP) -c "_64|d64")

The makefile is failing to set IS_X86 and IS_X86_64 properly under MinGW and its sh shell. When I add the following after the block of tests:

$(info SYSTEM: $(SYSTEM))
$(info IS_X86: $(IS_X86))
$(info IS_X86_64: $(IS_X86_64))

And I get in return:

$ make
IS_X86: 0
IS_X86_64: 0

I'm presuming MinGW can match i686 to i.86, so I'm guessing there's something wrong with the echo command or the variable under MinGW.

I also tried the following with no joy:

$(shell echo $(MACHINE) ...)
$(shell echo $((MACHINE)) ...)

I even tried to force the variable expansion before the tests:


Finally, I tried using a temporary with no joy:

MTEXT ?= $(shell $(UNAME) -m)

The tests work fine under the BSDs, Linux, OS X and Solaris. It even works under Cygwin.

What is the trick to using echo under MinGW to echo a makefile variable?

  • 1
    Perhaps, your make does not use the bash but another unexpected sh binary in your mingw environment ... Commented Jul 25, 2015 at 12:52
  • @maxschlepzig - you are right - its /bin/sh. Let me get it tagged properly...
    – user56041
    Commented Jul 25, 2015 at 12:57

1 Answer 1


It looks like you are using a weird shell in your mingw environment - and not bash. Or a weird version of egrep.

You can work around this via replacing those egrep shell calls with GNU make conditionals, e.g.:

ifeq (,$(filter %686 %x86 %i86 %amd64,$(MACHINE)))
  IS_X86 = 0
  IS_X86 = 1

Alternatively, you could create something based on $(findstring ...) or other GNU make text functions.

Besides working around your current problem, using such GNU make language feature is also more efficient than forking left and right via $(shell).

  • Thanks maxschlepzig. I'm not sure what shell it is. I'll try your recommendations. I keep Cygwin and MinGW around for testing only. We are getting ready to release Crypto++ 5.6.3, so I'm in a testing cycle.
    – user56041
    Commented Jul 25, 2015 at 14:23

You must log in to answer this question.