Hello Nigko. I believe that it is natural to think local-file were the
right procedure to use, and it could be used with relative path. Your
work-around puts in an absolute path, which perhaps really is what we
should put in the cookbook and guile-channel and blog post.
If a relative path is bad, we should warn against `local-file' with
relative paths in the manual and cookbook, and not just change the
example. Or yet rather make a change to `local-file' so calling it with
"../.." relative paths is treated right in all cases. Your diff makes
it right in more cases.
I believe `local-file' already is symlink-resolving in most cases:
Nigko Yerden <nigko.yer...@gmail.com> writes:
> pelzflorian (Florian Pelz) wrote:
>> But then again, the current current-source-directory
>> already does follow symlinks in nearly all cases, even in configuration
>> files
> Don't see it neither in the code nor in our examples. 'syntax-source' doesn't
> do this. The second branch of 'if' in 'absolute-dirname' bringing in
> 'canonicalize-path' is never executed in the examples (remember, the original
> patch didn't work because of this).
>
>> but actually I do not know why `local-file', when
>> calling `absolute-dirname', takes this case of `if'.
> Only the first branch of 'if' is executed in all practical cases I can
> imagine.
While processing guile-package.scm,
(search-path %load-path "guile-package.scm")
returns an absolute path if and only if guile-package.scm is in the
load-path, like when using it from a channel. Then, your diff makes it
resolve symlinks.
If the configuration or package file is not in the load-path,
guile-package.scm is returned, absolute-dirname’s other `if' branch
calls `canonicalize-path' on all but the basename and directory symlinks
already got resolved.
Do I misunderstand? I think symlinks are followed by design here.
(I wonder if such non-channel evaluation might cause problems when a
non-channel scheme file has the same name as a file in a channel. This
might be a rationale for never using local-file with relative paths.)
In a guix repl evaluating a file at a path with symlinks, containing a
call to the built-in pk procedure on (current-source-directory),
(let ((f (open-input-file "/tmp/a/b/c/a/b/c/d.scm")))
(eval (read f) (interaction-environment))
(eval (read f) (interaction-environment))
(eval (read f) (interaction-environment))
(eval (read f) (interaction-environment)) )
all unnecessary path components remain, no symlinks are resolved and
absolute-dirname is not called at all. This weird usage is unlike guix
home/system reconfigure or what normal people do.
>> Why does current-source-directory use syntax-source?
> What can it use instead? Related syntaxes such as
> 'current-source-location' and 'current-filename' are all using
> 'syntax-source' under the hood.
`current-module' from your work-around does not use syntax-source, but
outside modules cannot replace more powerful current-source-directory.
Regards,
Florian
