[Lxr-dev] Re: [ lxr-Bugs-455480 ] Diff. btwn. --allversions and --version=

[Jan-Benedict G. <jb...@lu...>]

On Thu, 2001-09-27 23:35:56 +0900, Malcolm Box <ma...@br....u=
k>
wrote in message <3BB...@br...>:
> In a bug report, Jan-Benedict wrote
>=20
> > There are these famous "LXR/xxx.pm not found" errors
> > around.
> >=20
> > They seem to depend on how the source tree (my one is
> > based
> > on CVS and PostgreSQL) is indexed.
> >
> > Indexing it with --allversions is quite fine: I can
> > browse the code
> > without any problems. But indexing specific versions
> > via
> > --version=3Dxxx (with xxx being a valid CVS tag) produces
> > those
> > "not found" errors afterwards. So it seems that genxref
> > either
> > fails to generate correct indexes for some given cvs
> > tag, or
> > "source" tries to get wrong releases from
> > cvs/database...
>=20
> Are you still experiencing this problem?  =18I can't see any reason why
> the state of the database would result in "not found" errors - it's
> perfectly possible to run lxr against an empty database, so I'm confused
> by this bug.

Well, I did a new installation (from current CVS) just right now
and here are my findings:

- lxr.css is not there. I've commented out anything using it:-)
- I indexed my CVS archive (using postgresql) version-by-version:

	for i in v_0_01 v_0_11 v_0_12 ...; do
		./genxref --url=3Dhttp:/... --version=3D$i
	done

Then I tried to access them via LXR:

	- It seems to not properly close connections to the DB server.
	  After viewing a page, I would expect apach to no longer
	  have any opened DB handles. But postgres thinks different
	  about that:
		mirror:~# ps ax|grep post
		17542 ?        S      0:00 /usr/lib/postgresql/bin/postmaster -D /var/lib=
/postgres/data
		23455 ?        S      0:04 postgres: www-data lxr-cvs [local] idle in tra=
nsaction     =20
		23457 ?        S      0:03 postgres: www-data lxr-cvs [local] idle in tra=
nsaction     =20
		23459 ?        S      0:02 postgres: www-data lxr-cvs [local] idle in tra=
nsaction     =20
		23499 ?        S      0:00 postgres: www-data lxr-cvs [local] idle in tra=
nsaction     =20
		23521 ?        S      0:01 postgres: www-data lxr-cvs [local] idle in tra=
nsaction     =20
		23524 ?        S      0:01 postgres: www-data lxr-cvs [local] idle in tra=
nsaction     =20
		23751 pts/5    S      0:00 grep post
	- I again needed to include this mysterious "lib" line to
	  source to make it work at all:
	  	use lib do { $0 =3D~ m{(.*)/} ? "$1/lib" : "lib" };
	- Not having done --allversions yet, I am not capable of
	  viewing that much pages. As the old behaviour was, after
	  6 pages (# of apache workers) I get the first "...not
	  found..." error messages.
	- I missed to look for zombies...
	- I shut apache down and indexed with --allversions
	- Restated apache and I "feel" (TM) that getting pages is
	  better, but not perfect. After some pages (but not that
	  fast as previously), I got my "not found" error.
	  Additionally, I find "co" zombies as being apache's childreen:
		mirror:~# ps ax|grep defun
		23654 ?        Z      0:00 [co <defunct>]
		23684 ?        Z      0:00 [co <defunct>]
		23732 ?        Z      0:00 [co <defunct>]
		23787 pts/5    S      0:00 grep defun
=09
> Here's a few questions that may help nail it down.  Firstly, what are
> the modules that are reported as "not found"?

I only saw LXR/Lang/Generic.pm in those messages...
=09
> If you index with --allversions, and then specify a version that doesn't
> exist, do you get the same "not found" messages as with --version?

If a version doesn't exist, I always get:

        The file /include/asm/memory.h does not exist.
** Warning: Unable to open /home/data/CVSROOT/linux//include/asm/memory.h,v

=2E..which is perfectly okay (for a non-indexed version of that file).

I have some other hint. I've got a very "special" construct in my
lxr.conf:

	'a'	=3D> {	'name'		=3D> 'Architecture',
			'range'		=3D> [qw(i386 alpha arm cris ia64 m68k mips mips64 parisc ppc p=
pc64 s390 s390x sh sparc sparc64 um x86_64 noarch)],
			'default'	=3D> 'noarch'
	},

and

	'maps'	=3D> {	'/include/asm-noarch/'	=3D> '/include/asm/',
			'/include/asm[^\/]*/'	=3D> '/include/asm-$a/',
			'/arch/[^\/]+/'		=3D> '/arch/$a/',
	},

I've tested around and found this "table" (it's called a hash,
isn't it?) internally gets processed in reverse order. So, IRL
"/asm/" would get substituted by "/asm-noarch/" which is to be
substituted again by "/asm/" having "noarch" the default (this is
for old linux kernel which only had an /include/asm/ directory
but not a *link* pointing to supported architecture's asm files...

I tried to comment out all maps and to default to '' for
architecture, but that didn't help: no files where found any longer.

MfG, JBG

--=20
Jan-Benedict Glaw . jb...@lu... . +49-172-7608481