Re: [Gnu-arch-users] [HEY TOM!] unhelpful error for unreadable directories

[Michael Poole]

Andre Majorel writes:

> How about:
>
>   E 4269 foo/{arch}: Permission denied
>
> In the man page, in the DIAGNOSTICS section:
>
>   4269 _filename_: _error_
>       An error occurred while lstating _filename_. If the error
>       is "Permission denied", verify that the parent directory
>       is searchable.
>
> Heresy ?

This scheme is popular with some vendors, since it provides an
identifier identifier for each error message that does not vary with
language.  This verges on being a FAQ for gcc, where the question
comes up about once a year.  One good explanation of why gcc does not
do this is given at http://gcc.gnu.org/ml/gcc/2002-03/msg00928.html

To summarize Joseph Myer's post:

1) It is of dubious value; if a user gets the output in the Elbonian
language, and you need the English version, you can either ask an
Elbonian-speaking developer to translate or grep for the translation.
If further dialogue is necessary to diagnose the problem, an
Elbonian-speaking developer is also the most qualified to do so.

2) The identifier-assigning authority must decide whether two errors
are the "same" or not for the purposes of diagnosis -- and more
importantly, whether they will remain the same.

One thing that he does not mention, but which is especially relevant
with and for arch, is:

3) When a project has branches, there is no central authority to
assign these unique identifiers unless you include the archive
location.  If you do that, the main benefit (brevity) is lost.

And there's always this one:

4) Cryptic numbers and strings are intimidating, and the gist of this
discussion is how to make error messages more useful to users.

Michael