[Ada-mode-users] Usage of compilation-search-path considered harmful

[Ludovic Brenta]

Hello,

compile.el contains:

(defcustom compilation-search-path '(nil)
  "List of directories to search for source files named in error 
messages.
Elements should be directory names, not file names of directories.
The value nil as an element means to try the default directory."
  :type '(repeat (choice (const :tag "Default" nil)
                         (string :tag "Directory")))
  :group 'compilation)

As can be seen, the intention is to navigate from a compilation buffer 
to
a source file, not between source files. Therefore, it is my opinion 
that
this variable should always be buffer-local to a compilation buffer.  
Also,
the fact that this variable is defined by "defcustom" implies that it 
is
intended to be under the sole control of the human user, by means of
customize-variables.

I think ada-mode violates both of these principles and has done so for 
many
years.  Grepping in emacs 24.5, I see that only ada-mode 4.00 ever 
mentions
compilation-search-path; I have not investigated why.  No other 
programming
mode does that.

ada-mode 5.2.1 sets this variable when parsing a project file (either 
.gpr
or .adp).  This has the following effects:

* when a user does M-x compile, whatever compilation results in the 
buffer
  will use compilation-search-path, even if the compilation has nothing 
to
  do with the ada-prj-current-file.  The user may start an arbitrary
  compilation command, e.g. to compile some COBOL files :)

* when a user does M-x grep, this opens a *grep* buffer which inherits 
from
  compilation-mode and therefore uses compilation-search-path.  This is
  harmful because the user controls where grep runs and certainly does 
not
  want files in the compilation-search-path to hide ones in the grep 
results.

* when a user does M-x occur, this also opens a *occur* buffer which 
inherits
  from compilation-mode, so the same remark applies.

After lengthy discussions with my colleagues and some more lengthy 
meditation
on this, I have reached the conclusion that is is best that ada-mode 
*not*
use, mention or change compilation-search-path at all.  Navigation from 
one
source buffer to another (e.g. ada-goto-declaration) uses other means, 
namely
a gnatfind buffer; and navigation from a compilation buffer to source 
should
use a buffer-local value in the compilation buffer.  If 
compilation-search-
path is nil, emacs has a reasonable default for this.

A first glance at the usage of compilation-search-path in ada-mode 
5.2.1
tells me that there is no compelling reason for it.

Opinions anyone?

--
Ludovic Brenta.