On 2016-07-05T15:31:57-0300, Marc Collin wrote:
> I did some tests with 8cc (because it's easier to build than scc for me).
> My results are that the binaries aren't identical.
> Here's a script so you can try to reproduce it.
>
> #!/bin/sh
> mkdir test_comp
> cd test_comp
> git clone https://github.com/rui314/8cc
> mv 8cc 1c
> cp -r 1c 2c
> cp -r 1c Ac
> cp -r 1c Bc
>
> # compile 8cc with gcc
> cd 1c
> make cc=gcc
>
> # compile 8cc with clang
> cd ../2c
> make cc=clang
I don't think this does what you expect it to do. At least looking at
the 8cc makefile, the compiler is hardcoded as `cc', not the expansion
of the variable `cc'. Try
make cc=/bin/false
or such.
> # compile 8cc with 8cc
> # using the binary that
> # was compiled with gcc
> cd ../Ac
> make cc=../1c/8cc
>
> # compile 8cc with 8cc
> # using the binary that
> # was compiled with clang
> cd ../Bc
> make cc=../2c/8cc
>
> cd ..
>
> # compare binaries
> md5sum Ac/8cc Bc/8cc
>
On my system, the resulting binaries include references to the path they
were compiled at.
$ ls */8cc
8cc/8cc other_8cc/8cc
$ strings 8cc/8cc | grep home
/home/sgilles/temp/a/8cc/include
#include </home/sgilles/temp/a/8cc/include/8cc.h>
/home/sgilles/temp/a/8cc
$ strings other_8cc/8cc | grep home
/home/sgilles/temp/a/other_8cc/include
#include </home/sgilles/temp/a/other_8cc/include/8cc.h>
/home/sgilles/temp/a/other_8cc
I suspect this is what actually causes the differences in md5sums.
--
S. Gilles
