Re: Adding 'compat' from ELPA as an optional dependency to ERC

[Philip Kaludercic <philipk@posteo.net>]

"J.P." <jp@neverwas.me> writes:

> Hi Philip (and Stefan),
>
> Philip Kaludercic <philipk@posteo.net> writes:
>
>>                                             By explicitly loading
>> compat-28, one would additionally load "prefixed" definitions for
>> functions that add new features to existing functions.
>>
>> E.g. consider how assoc acquired a new optional argument "TESTFN" in
>> version 26.  One would either have to advise the existing function,
>> which was regarded to be too invasive and dangerous, or to define a
>> separate function underneath the `compat-' prefix -- `compat-assoc'.
>>
>> This function is not made visible by loading `compat', and explicitly
>> ought not to be in this case, as someone using the in-tree version of
>> ERC would not have no such function defined.  In this sense the support
>> compat can provide for core packages is limited, but as it seems for
>> ERC's requirements sufficient.
>
> I suppose we could always add prefixed definitions down the road if
> really needed, perhaps to spare us from having to copy some complicated
> backport definition in full. 

There is always the possibility of changing compat to advise certain
functions instead of providing prefixed alternatives.  But that should
probably be decided in a case-by-case basis.

Depending on the interest of the Emacs maintainers in `compat' as a
project, it might also be possible to add compat-* definitions to the
core, thought I am not certain sure how fond I am of this idea.

>                              
>                              Although, I believe doing so would also
> define compat- prefixed variants in all versions of Emacs, the idea
> being to save third-party packages from having to do stuff like
>
>   (defmacro erc-compat--json-parse-string (string &rest args)
>     `(,(if (fboundp 'compat-json-parse-string)
>            'compat-json-parse-string
>          'json-parse-string)
>       ,string
>       ,@args))

If a function like this were to be definied upstream

        (defun get-compatibility-func (name)
          "Return the function NAME of a compatibility alias."
          (let* ((compat (intern-soft (format "compat-%s" name))))
            (or (symbol-function (if (fboundp name) name compat))
                (error "No definition for %S could be found" name))))

then all that ERC would have to do is

        (defalias 'erc-json-parse-string
          (get-compatibility-func 'json-parse-string))

> But in ERC's case, there'd be no avoiding this kind of indirection.
> (Please correct me if that's a flawed understanding.)

While prefixing is used, no, because we cannot assume that compat is
installed.