Aaron J. Grier
aaron at frye.com
Wed Jul 30 18:28:57 UTC 2003
On Wed, Jul 30, 2003 at 08:17:09AM -0400, Ed Sutter wrote:
> 1. The output of the --line-numbers option to NM does not build the
> relative paths properly (not sure why). For example, one entry in the
> output may be:
> but the preceding "../../../../.." may not be right.
this is dependent on how you run configure. if you always specify a
full path when running configure from your build directory, things work
out a little better. so instead of ../../../../foo/bar/blatz you'll get
/usr/local/build/quux/../../../foo/bar/blatz. it's still not terribly
readable, but gdb will have an easier time finding source files
without having to add a bunch of directory searches.
> 2. This only includes the files that actually originate symbols, so no
> header files are in this list. One crude way to add headers is to
> just use find...
> find BASEDIR -name '*.h' >>cscope.files
> (where BASEDIR is top level of your RTEMS tree).
> Bad thing about this is that it pulls in ALL header files, so the
> output must be parsed. A more accurate list of headers could
> probably be derived from the output of "gcc -MM" (used to build a
> file dependency list), but I haven't done that yet.
in general it'd be nice to have cross-directory dependencies in the
RTEMS tree. I don't know if this has improved in 4.6.0, but in 4.5.0
changing a device driver header in a BSP won't necessarily force a
recompile of a test program which includes that header. ideally I'd
like to see a non-recursive make, but that's quite a bit of work. (:
and don't forget to add the inline files (.inl) and the
autotools-generated .h files (.h.in) to cscope's file list.
Aaron J. Grier | Frye Electronics, Tigard, OR | aaron at frye.com
More information about the users