[GRASS-dev] clean_temp rewritten

[Markus Neteler]

Hi,

I have (after a while of testing) submitted the new
lib/init/clean_temp.c code from Roberto Flor (ITC-irst).

The new code has better timing to remove old leftover
tmp files and also removed leftover subdirectories now
(in LOCATION/MAPSET/.tmp/).

Here some related offlist discussion excepts from the
test phase of the new version:

On Tue, May 16, 2006 at 04:20:43AM +0100, Glynn Clements wrote:
> Markus Neteler wrote:
> > Markus Neteler wrote:
> > > Glynn Clements wrote:
> > > >         if ( (pathlen=snprintf(buf,BUF_MAX,"%s/%s",pathname,cur_entry->d_name)) >= BUF_MAX)
> > > >                 G_fatal_error("clean_temp: exceeded maximum pathname length %d, got %d, should'nt happen",BUF_MAX,pathlen);
> > >
> > > Unfortunately:
> > >
> > > 1. snprintf() is C99; not all systems have it.
> >
> > So far it is implemented a few times in GRASS:
> >
> > find . -type f -name "*.c" -exec grep -l snprintf {} \;
> > ./raster3d/r3.in.ascii/main.c
> > ./db/drivers/dbf/dbfexe.c
> > ./raster/r.support/front/front.c
> > ./raster/r.support/front/check.c
> > ./raster/r.support/front/run.c
> > ./raster/r.support/modhead/check_un.c
> > ./raster/r.support/modhead/modhead.c
> > ./raster/r.support/modhead/ask_format.c
> > ./lib/db/sqlp/lex.yy.c
> > ./lib/db/dbmi_client/select.c
> > ./lib/gis/user_config.c
> > ./lib/vector/dglib/examples/components.c
> >
> > Should this become at least a G_snprintf() wrapper
> > in libgis?
>
> I'm not sure.
>
> If libc doesn't provide snprintf, the only options are to:
>
> 1. Implement it from scratch.
> 2. Write a wrapper which uses fprintf() to write the string to a
> temporary file then reads the temporary file into the buffer.
> 3. Write a wrapper which discards the size argument and calls
> vsprintf().
>
> Option 1 is a fair amount of work (even extracting the code from libc
> isn't simple). Option 2 is messy and option 3 creates the risk of
> buffer overruns.
>
> This issue has been discussed on the list in the past; IIRC, the
> discussion was initiated by someone trying to compile GRASS on a
> system which lacked snprintf(). I don't recall the conclusion.

This was also Glynn's suggestion in 2002 AFAIK:
 http://grass.itc.it/pipermail/grass5/2002-April/005218.html
 "Should I add a stub implementation to libgis?"

Here is the full thread:
 http://grass.itc.it/pipermail/grass5/2002-April/thread.html#5117
 "snprintf and other compile errors on IRIX"

IMHO, a quick solution to the problem is:
- make G_snprintf() a wrapper around snprintf()
- test therein if the system support snprintf(), if not, issue
  an error (or write an implementation, see below).

In this case we could identify systems with snprintf() being
unsupported using G_snprintf() and add extra tricks.

BTW, here a thread on mingW:
 http://lists-archives.org/mingw-users/03714-mingw-snprintf-not-following-c99-standard.html

See here for a GPL'ed snprintf.c (suggested in above mingW thread):
 http://www.ijs.si/software/snprintf/


Markus