Re: Nonsensical byte compiler warning.

[Chong Yidong]

Markus Triska <address@hidden> writes:

> Alan Mackenzie <address@hidden> writes:
>
>> Would it be easy to make this change to the byte compiler?
>
> Alas, no. The optimiser hasn't got the positioning information, and
> the byte-code emitter potentially doesn't see such calls at all:
> Conceptually, the optimiser is capable of totally removing unneeded
> function calls (the code is there, and disabled). Conversely, the
> positioning code is fragile enough already to warrant keeping it well
> separated from the optimiser, and limited to the emitter. Now, since
> unnecessary function calls are currently kept, detection logic from
> the optimiser could be used in the emitter, but that would duplicate a
> lot of code, is error-prone, and can, in my view, not yield more
> useful results than byte-compiling with byte-optimize-log set to t. On
> top of that, it would result in the same char-before/after confusion,
> since that rewriting is performed by the optimiser, not the emitter.

If the offending function name is unreliable, maybe we should omit it
and issue a "value returned from form is not used" warning.  What do
you think?