getting rid of mkproto.sh from Samba3
Stefan (metze) Metzmacher
metze at samba.org
Wed Jun 6 10:35:53 GMT 2007
-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1
James Peach schrieb:
> On 05/06/2007, at 5:10 PM, tridge at samba.org wrote:
>
>> James,
>>
>>> If I have a decent header file, I can see the full interface in a
>>> single screen, along with comments, macros and typedefs. I don't think
>>> the argument about the extra typing involved in changing headers by
>>> hand is very strong, since the code needs to be read and understood
>>> thousands more times than it needs to be written.
>>
>> It is not the extra typing, it's the mistakes. Before we had auto
>> prototypes in Samba the following mistakes were common:
>>
>> *) leaving a prototype in a header for a function that has been
>> removed. I don't know of any tool that will find/remove those.
>>
>> *) getting details in the prototype wrong, especially the varargs
>> functions with printf arguments.
>>
>> *) when the arguments to a function change, and the prototype isn't
>> updated. This is especially insidious when the new and old arguments
>> are type compatible, but change their meaning, so the old prototype
>> is completely misleading (ie. is still takes an int as the 3rd
>> argument, but now its a set of flags, not a counter)
>>
>> *) once you start breaking up your includes out of a single
>> includes.h, then you risk getting the prototype completely wrong if
>> the source file of the function doesn't include the prototype
>> header. The compiler won't complain.
>>
>> *) programmers add nice comments to the prototype header, which is in
>> a completely different file to the code. Those comments will degrade
>> over time as the code changes and the comment is not upgraded, so
>> those nice docs that you do this month will become completely wrong
>> by next year, and the person updating the function will not notice,
>> as the prototype is not on the same screen as the code they are
>> changing.
>
> OK, these are pretty convincing. I'll take a look at integrating the
> Samba4 mkproto, that seems uncontroversial and likely to improve things.
I think the model we have in samba4 is really nice:
we have per subsystem private and public headers generated
But I think we can do better if we commit the public one's to svn
and don't generate them by the normal 'make'.
To keep information only in one place we should improve mkproto.pl
to also copy doxxygen comments into the proto header.
Then we should add a the need of on exlicit 'make
regenerate_public_headers', which will print out warnings about
changed headers.
I think this will solve most of the problems. It's impossible
to change public functions without human interaction and
private functions are as simple as they're now.
Also we will have uptodate comments in the headers without
risking cut-n-paste or laziness bugs of manually keep
source files and headers in sync.
I had also planed to a some diagnostic checks into the samba4 build
system, to automaticly check that the subsystem public header (
not only the public proto header) is self containing with the specified
public dependencies.
And it'll also check a subsystem links fine only using it's public and
private dependencies.
metze
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.5 (GNU/Linux)
Comment: Using GnuPG with SUSE - http://enigmail.mozdev.org
iD8DBQFGZo4Jm70gjA5TCD8RAuiyAJ940umru00MiZXKD81FuMj3eWPE+wCg0koI
X2Qb1X8f1SSrxiEYy6E6lT0=
=tSQt
-----END PGP SIGNATURE-----
More information about the samba-technical
mailing list