cluts makefiles

cluts makefiles

[Vasiliy Kulikov]

Hi,

This is a patch to enhance musl building things.

1) Divided a single Makefile to the cluts, tests/, compile flags.

2) Used gcc's ability to identify dependencies.


Unrelated things:

include "sequence.c" is weird :)  It's better to use .h with
declarations and .c files which build into .o.

With glibc:

a) _SVID_SOURCE is needed for alphasort.

b) SA_NODEFER is undefined in all .c.


diff --git a/Makefile b/Makefile
index 07377a8..e77ca3a 100644
--- a/Makefile
+++ b/Makefile
@@ -7,18 +7,27 @@
 # There's ABSOLUTELY NO WARRANTY, express or implied.
 #
 
-CFLAGS = -std=c99 -D_POSIX_C_SOURCE=200809L -Wall -Wextra -O2
-LIBS = -lpthread -lrt -lm
-SRC = $(wildcard tests/*.c) $(wildcard *.c)
-BIN = $(SRC:.c=)
+include defines.mk
 
-%:%.c
+SRC = cluts.c
+BIN = cluts
+
+all: TESTS $(BIN)
+	@:
+
+%.o: %.c
+	$(CC) $(CFLAGS) $(LIBS) -c $<
+
+%: %.o
 	$(CC) $(CFLAGS) $(LIBS) -o $@ $<
-all: $(BIN)
+
+TESTS:
+	cd tests && make
 
 %.run:%
 	./$<
 test: all $(SRC:.c=.run)
 	
 clean:
-	rm -f $(BIN)
+	rm -f $(BIN) *.o
+	cd tests && make clean
diff --git a/defines.mk b/defines.mk
new file mode 100644
index 0000000..220db82
--- /dev/null
+++ b/defines.mk
@@ -0,0 +1,2 @@
+CFLAGS = -std=c99 -D_POSIX_C_SOURCE=200809L -Wall -Wextra -O2
+LIBS = -lpthread -lrt -lm
diff --git a/tests/Makefile b/tests/Makefile
new file mode 100644
index 0000000..c6c3b5e
--- /dev/null
+++ b/tests/Makefile
@@ -0,0 +1,25 @@
+include ../include.mk
+
+SRC = $(wildcard *.c)
+BIN = $(SRC:.c=)
+
+all: COMMON $(BIN)
+	@:
+COMMON:
+	@:
+
+%.o:
+	$(CC) $(CFLAGS) $(LIBS) -c $<
+
+%: %.o
+	$(CC) $(CFLAGS) $(LIBS) -o $@ $<
+
+deps.mk: $(SRC)
+	$(CC) -MM $^ > $@
+
+ifneq ($(MAKECMDGOALS),clean)
+-include deps.mk
+endif
+
+clean:
+	rm -rf $(BIN) *.o
---

Re: cluts makefiles

[Luka Marčetić]

On 08/09/2011 01:30 PM, Vasiliy Kulikov wrote:
> Hi,
>
> This is a patch to enhance musl building things.
>
> 1) Divided a single Makefile to the cluts, tests/, compile flags.
>
> 2) Used gcc's ability to identify dependencies.

Do others free compilers support nr 2?

> Unrelated things:
>
> include "sequence.c" is weird :)  It's better to use .h with
> declarations and .c files which build into .o.

It's an ad-hoc dir structure, therefore suboptimal. I'll add prototypes 
and other declarations into .h some day.

> With glibc:
>
> a) _SVID_SOURCE is needed for alphasort.

Maybe older glibc needed it (I get no warnings). I'm not sure it 
warrants an inclusion (it's SUSv4).

> b) SA_NODEFER is undefined in all .c.

It should be defined in signal.h. Alexander had problems with this. 
Trying to apply the patch that you wrote told me you're missing an 
_XOPEN_SOURCE flag I've relatively recently added to the Make-file on 
Rich's incentive, so perhaps you could should try that. If it's an old 
kernel version instead, necessary modifications would be more extensive, 
involving unblocking when entering the handler (sigprocmask?), removing 
SA_NODEFERs, using sig*jmp, and perhaps more in some cases.

> diff --git a/Makefile b/Makefile

Having made dirs a and b (both from the same repo dir), I tried `patch 
-p0 < yourpatch`, and got:

Hunk #1 FAILED at 7.
1 out of 1 hunk FAILED -- saving rejects to file a/Makefile.rej
patching file b/defines.mk
patching file b/tests/Makefile

I tried to replace the offending line with:

-CFLAGS = -std=c99 -D_XOPEN_SOURCE=700 -D_POSIX_C_SOURCE=200809L -Wall 
-Wextra -O2

but somehow the error message persists. Is there something I'm missing?

Thanks,
Luka

Re: cluts makefiles

[Vasiliy Kulikov]

On Tue, Aug 09, 2011 at 17:48 +0200, Luka Marčetić wrote:
> On 08/09/2011 01:30 PM, Vasiliy Kulikov wrote:
> >2) Used gcc's ability to identify dependencies.
> 
> Do others free compilers support nr 2?

Hmm, not sure.  I think it is gcc specific thing.  Then you could use it
if there is gcc installed (just installed, CC can be arbitrary
compiler) and leave deps.mk blank as a fallback and change target line
from

%.o:

to

%.o: %.c

Or you could precompile these files with gcc and distribute them as is
without gcc dependency.


> >b) SA_NODEFER is undefined in all .c.
> 
> It should be defined in signal.h. Alexander had problems with this.

The target system is Ubuntu 10.04, glibc-2.10-1, gcc 4.4.3-4ubuntu5.


> >diff --git a/Makefile b/Makefile
> 
> Having made dirs a and b (both from the same repo dir), I tried
> `patch -p0 < yourpatch`, and got:
> 
> Hunk #1 FAILED at 7.
> 1 out of 1 hunk FAILED -- saving rejects to file a/Makefile.rej
> patching file b/defines.mk
> patching file b/tests/Makefile

You should use "-p1".  With -p0 patch doesn't remove any prefix
directories and tries to change files b/defines.mk and similar.

-- 
Vasiliy

Re: cluts makefiles

[Luka Marčetić]

On 08/09/2011 06:11 PM, Vasiliy Kulikov wrote:
>> Hunk #1 FAILED at 7.
>> 1 out of 1 hunk FAILED -- saving rejects to file a/Makefile.rej
> You should use "-p1".  With -p0 patch doesn't remove any prefix
> directories and tries to change files b/defines.mk and similar.

Where should this be run? I tried root where a/ and b/ are, and also in 
the b/ itself. In both cases, the same thing happens. Maybe it is me who 
messed up line 7 now, I don't know. :-/

>>> 2) Used gcc's ability to identify dependencies.
>> Do others free compilers support nr 2?
> Hmm, not sure.  I think it is gcc specific thing.

Ideally, I'd use the standard make, but I'm a make noob, so I use the 
GNU one. However, I would deprecate a dependency on gcc - I want my C99 
code to be compilable with any C99-compliant compiler. If you manage to 
remove the dependency on gcc (and instruct me how to apply the patch), 
I'll certainly prefer the right(tm) way to build things.

>>> b) SA_NODEFER is undefined in all .c.
>> It should be defined in signal.h. Alexander had problems with this.
> The target system is Ubuntu 10.04, glibc-2.10-1, gcc 4.4.3-4ubuntu5.

Granted, I use kinda beta software (Debian Sid here), but I don't think 
that's it. Did you try the POSIX flag?
Thanks again.
Luka

Re: cluts makefiles

[Solar Designer]

Luka -

On Tue, Aug 09, 2011 at 10:20:24PM +0200, Luka Mar??eti?? wrote:
> On 08/09/2011 06:11 PM, Vasiliy Kulikov wrote:
> >>Hunk #1 FAILED at 7.
> >>1 out of 1 hunk FAILED -- saving rejects to file a/Makefile.rej
> >You should use "-p1".  With -p0 patch doesn't remove any prefix
> >directories and tries to change files b/defines.mk and similar.
> 
> Where should this be run? I tried root where a/ and b/ are, and also in 
> the b/ itself. In both cases, the same thing happens. Maybe it is me who 
> messed up line 7 now, I don't know. :-/

To apply a patch, you do not need two trees, and you do not need to
match the top-level directory names to those seen in the patch file.
Rather, you may do something like:

cp -a cluts cluts-to-patch
cd cluts-to-patch
patch -p1 < ../patch-file

or just:

cd cluts
patch -p1 < ../patch-file

but the first approach may be preferable if you haven't yet made the
determination to fully apply the patch to your tree (so you want to
experiment on a copy first).

Here are some instructions/examples on how to generate patches and how
to apply them:

http://openwall.info/wiki/how-to-make-patches
http://openwall.info/wiki/john/how-to-extract-tarballs-and-apply-patches

You may want to experiment with this for your own changes to cluts -
just so that you know how it works - before you apply a patch provided
to you by someone else.

Alexander

Re: cluts makefiles

[Luka Marčetić]

On 08/10/2011 05:06 AM, Solar Designer wrote:
> To apply a patch, you do not need two trees, and you do not need to
> match the top-level directory names to those seen in the patch file.
> Rather, you may do something like:
>
> cp -a cluts cluts-to-patch
> cd cluts-to-patch
> patch -p1<  ../patch-file

Hmm, I did try`patch  -p1 < ../patch-file` *shrugs*
Still, thanks you for the instructions anyway. At least I can be sure 
what I'm doing.
Luka

Re: cluts makefiles

[Solar Designer]

On Wed, Aug 10, 2011 at 01:29:31PM +0200, Luka Mar??eti?? wrote:
> On 08/10/2011 05:06 AM, Solar Designer wrote:
> >To apply a patch, you do not need two trees, and you do not need to
> >match the top-level directory names to those seen in the patch file.
> >Rather, you may do something like:
> >
> >cp -a cluts cluts-to-patch
> >cd cluts-to-patch
> >patch -p1<  ../patch-file
> 
> Hmm, I did try`patch  -p1 < ../patch-file` *shrugs*
> Still, thanks you for the instructions anyway. At least I can be sure 
> what I'm doing.

Well, it sounds like your source tree was in fact changed from the
revision that Vasiliy's patch was for, then.  Such things happen.

Alexander