some numbers on macros

some numbers on macros

[Mariusz Kozlowski]

[-- Attachment #1: Type: text/plain, Size: 3258 bytes --]

Hello,

	I wrote a rather dumb script to see some numbers on macros I was
interested in. The script basically parses *.c file, finds macro definitions
and counts how many times each macro was used in this source file. The script
doesn't see any context so it can produce false positives - that is one of
the reasons it doesn't look into header files - it's simply too dumb.

The macros I was interested in were these which are:
- defined and unused
- defined and used only once
- defined more than once

I dont want to start another cleanup-patches-storm so first I'd like to
know if this excersise is worth it as unused macros do not do much harm.

Would such patches be accepted?

	Mariusz



example on recent linux-2.6 snapshot:

$ ./find_interesting_macros.php linux-2.6
[snip]
10189 files to parse

used  defined   macro
[snip]
0       1       'Tracev'                        /home/mariusz/linux/linux-2.6/init/initramfs.c +400
0       1       'Tracevv'                       /home/mariusz/linux/linux-2.6/init/initramfs.c +401
0       1       'Tracec'                        /home/mariusz/linux/linux-2.6/init/initramfs.c +402
0       1       'Tracecv'                       /home/mariusz/linux/linux-2.6/init/initramfs.c +403
1       1       'version_string'                /home/mariusz/linux/linux-2.6/init/version.c +17
8       2       'DBG'                           /home/mariusz/linux/linux-2.6/sound/soc/at91/at91-ssc.c +39
4       2       'DBG'                           /home/mariusz/linux/linux-2.6/sound/soc/at91/at91-pcm.c +37
8       2       'DBG'                           /home/mariusz/linux/linux-2.6/sound/soc/at91/eti_b1_wm8731.c +44
25      2       'DBG'                           /home/mariusz/linux/linux-2.6/sound/soc/s3c24xx/s3c2412-i2s.c +49
21      2       'DBG'                           /home/mariusz/linux/linux-2.6/sound/soc/s3c24xx/s3c24xx-i2s.c +49
19      2       'DBG'                           /home/mariusz/linux/linux-2.6/sound/soc/s3c24xx/s3c24xx-pcm.c +42
0       2       'dbg'                           /home/mariusz/linux/linux-2.6/sound/soc/codecs/wm8750.c +41
1       1       'info'                          /home/mariusz/linux/linux-2.6/sound/soc/codecs/wm8750.c +48
0       1       'warn'                          /home/mariusz/linux/linux-2.6/sound/soc/codecs/wm8750.c +50
1       1       'wm8750_reset'                  /home/mariusz/linux/linux-2.6/sound/soc/codecs/wm8750.c +120
0       2       'dbg'                           /home/mariusz/linux/linux-2.6/sound/soc/codecs/wm8731.c +41
1       1       'info'                          /home/mariusz/linux/linux-2.6/sound/soc/codecs/wm8731.c +48
0       1       'warn'                          /home/mariusz/linux/linux-2.6/sound/soc/codecs/wm8731.c +50
1       1       'wm8731_reset'                  /home/mariusz/linux/linux-2.6/sound/soc/codecs/wm8731.c +120
...

BTW. How many unused macros are in *.c files?

$ ./find_interesting_macros.php linux-2.6 | grep "^0" | wc -l
1340

Well ... since it produces some false positives the real number can differ a little.
This is an example of a macro that will cause false positive:

scripts/kconfig/kconfig_load.c +7
#define P(name,type,arg)        type (*name ## _p) arg
#include "lkc_proto.h"
#undef P


[-- Attachment #2: find_interesting_macros.php --]
[-- Type: application/x-php, Size: 3894 bytes --]

Re: some numbers on macros

[Arjan van de Ven]

On Sat, 24 May 2008 17:48:19 +0200
Mariusz Kozlowski <m.kozlowski@tuxland.pl> wrote:

> Hello,
> 
> 	I wrote a rather dumb script to see some numbers on macros I
> was interested in. The script basically parses *.c file, finds macro
> definitions and counts how many times each macro was used in this
> source file. The script doesn't see any context so it can produce
> false positives - that is one of the reasons it doesn't look into
> header files - it's simply too dumb.
> 
> The macros I was interested in were these which are:
> - defined and unused
> - defined and used only once
> - defined more than once
> 
> I dont want to start another cleanup-patches-storm so first I'd like
> to know if this excersise is worth it as unused macros do not do much
> harm.
> 
> Would such patches be accepted?

An unused define/macro that was declared in a .c file isn't very
interesting to be honest; that tends to be helper macros for the
developer and removing those is just counter productive.

unused macros in common header files are an indication of stale APIs
otoh.. and might be of some interest. (Same for static inline in
headers)

Re: some numbers on macros

[Matthew Wilcox]

On Sat, May 24, 2008 at 09:33:50AM -0700, Arjan van de Ven wrote:
> unused macros in common header files are an indication of stale APIs
> otoh.. and might be of some interest. (Same for static inline in
> headers)

Possibly.  There's likely to be a lot of macros unused like these ones:

#define  PCI_X_CMD_MAX_READ     0x000c  /* Max Memory Read Byte Count */
                                /* Max # of outstanding split transactions */
#define  PCI_X_CMD_SPLIT_1      0x0000  /* Max 1 */
#define  PCI_X_CMD_SPLIT_2      0x0010  /* Max 2 */
#define  PCI_X_CMD_SPLIT_3      0x0020  /* Max 3 */
...

where the macros embody the PCI specification in code -- possibly we
don't use them yet, but if we ever did, we'd have the macros to use.

Another category of false positive is macros that ought to be used,
but the code that ought to be using them has decided to go its own way.
That still indicates a bug, but not the one you might initially think.

In summary, this is probably an interesting exercise, but the results
would need to be interpreted with care.

-- 
Intel are signing my paycheques ... these opinions are still mine
"Bill, look, we understand that you're interested in selling us this
operating system, but compare it to ours.  We can't possibly take such
a retrograde step."

Re: some numbers on macros

[Mariusz Kozlowski]

Hello,

> On Sat, May 24, 2008 at 09:33:50AM -0700, Arjan van de Ven wrote:
> > unused macros in common header files are an indication of stale APIs
> > otoh.. and might be of some interest. (Same for static inline in
> > headers)
> 
> Possibly.  There's likely to be a lot of macros unused like these ones:

Maybe I wasn't precise enough. The script looks for macros which take argument(s)
i.e.

#define foobar() ...

> #define  PCI_X_CMD_MAX_READ     0x000c  /* Max Memory Read Byte Count */
>                                 /* Max # of outstanding split transactions */
> #define  PCI_X_CMD_SPLIT_1      0x0000  /* Max 1 */
> #define  PCI_X_CMD_SPLIT_2      0x0010  /* Max 2 */
> #define  PCI_X_CMD_SPLIT_3      0x0020  /* Max 3 */
> ...
> 
> where the macros embody the PCI specification in code -- possibly we
> don't use them yet, but if we ever did, we'd have the macros to use.

Agreed.

> Another category of false positive is macros that ought to be used,
> but the code that ought to be using them has decided to go its own way.
> That still indicates a bug, but not the one you might initially think.

Agreed.

> In summary, this is probably an interesting exercise, but the results
> would need to be interpreted with care.

Well I'll just go through some of them. This is long term task as its number is quite
big. Maybe I can improve the script some more or apply some other measures. I guess
one can find there all sort of macros and some part of them are leftovers from something
that was there some time ago.

Interesting part would be to write a 'parser' that actually sees the context and understands
cpp directives, that could 'walk' the tree.

	Mariusz

Re: some numbers on macros

[Arjan van de Ven]

On Sun, 25 May 2008 01:41:09 +0200
Mariusz Kozlowski <m.kozlowski@tuxland.pl> wrote:
> 
> Interesting part would be to write a 'parser' that actually sees the
> context and understands cpp directives, that could 'walk' the tree.


you could approach it from the opposite angle... I mean, lxr.linux.no
already has all this data in some database.. maybe you can turn it into
a set of database queries ;)

Re: some numbers on macros

[Mariusz Kozlowski]

Hello,

> On Sun, 25 May 2008 01:41:09 +0200
> Mariusz Kozlowski <m.kozlowski@tuxland.pl> wrote:
> > 
> > Interesting part would be to write a 'parser' that actually sees the
> > context and understands cpp directives, that could 'walk' the tree.
> 
> you could approach it from the opposite angle... I mean, lxr.linux.no
> already has all this data in some database.. maybe you can turn it into
> a set of database queries ;)

Good point :) Will look into this.

	Mariusz