Re: [Tack-devel] new PowerPC tests: ack-ptest

[George K. <ke...@gm...>]

On Wed, Nov 9, 2016 at 11:26 AM, Nelson H. F. Beebe <be...@ma...> wrote:
> However, before it can receive wide testing and use, ack really needs
> a new top-level main program that presents a POSIX-compatible compiler
> interface...

I believe that ack -E needs to work, ack -lm needs to work. Right now,
they don't work. POSIX seems to define both -E and -lm. Also, ack's
own build system uses both -E and -lm, so it must be impossible now
for ack to compile itself.

Current ack ignores unknown options (including -E). I believe that ack
needs to reject unknown options with an error message.

> I attempted a build of the snapshot on a freshly-installed Mac OS X
> 10.11.6 (El Capitan) system, where the default C compiler is named
> gcc, but is really clang under the hood.  The build fails...

Thank you for trying! So far I have not heard that ack can run on OS X.

I forgot to say that I don't run ack on OS X. I use my kernigh-osx
branch to cross-compile binaries for OS X from my machine running
OpenBSD/amd64.

Manoël Trapier tried to run ack on OS X in 2013. He doesn't seem to
have had much success. His code is in the to_ansi branch of
https://github.com/Godzil/ack

Because clang hates to compile old C code, the branch tried to convert
the code to ANSI C, by declaring return types and prototypes for
functions. My own PowerPC system with Mac OS X 10.4.11 and Xcode 2.5
is so old that it has an authentic gcc, not clang. I might have a
better chance of a successful compile.

But there's another problem: some ack tools use brk() and sbrk() to
allocate memory. Apple's brk() and sbrk() didn't work, so Manoël
Trapier provided (in modules/src/sbrk/sbrk_emu.c) an emulation of
brk() and sbrk() around calloc(), based on an example from D A Gwyn. I
now have a different emulation of brk() and sbrk() in
plat/osx/libsys/brk.c of my kernigh-osx branch, where ack
cross-compiles to OS X. It can't link with Apple's libraries, so I
must provide my own brk() and sbrk() for use with ack libc. I emulate
them around mmap() and mprotect().

> Build attempts were unsuccessful on Ubuntu 16.04.1 LTS (Xenial Xerus)
> and TrueOS (a FreeBSD 12 offshoot).  During those builds, I got
> numerous complaints like this:
>
>         Circular /tmp/ack-build/build.make <- /tmp/ack-build/build.make dependency dropped.
>
> and
>
>         Makefile:81: target 'w' given more than once in the same rule
>         Makefile:81: target '-f' given more than once in the same rule
>         Makefile:81: target '/tmp/ack-build/build.make' given more than once in the same rule

This might be a problem with ackbuilder's back end for make. I use the
back end for ninja because ackbuilder detects that my machine has
ninja. So my ack-build/ contains build.ninja, not build.make. I might
not install ninja on every machine where I try to build ack, so I
might encounter this problem later.

--George Koehler