Improve handling of `strerror` messages

[JayFoxRox]

(This is a low priority issue, but ports would probably require updates, so it's good to fix this before a bad implementation propagates)

As also indicated by the comment, this function is not ideal:

pdclib/functions/string/strerror.c

Lines 13 to 24 in 862f106

	/* TODO: Doing this via a static array is not the way to do it. */
	char * strerror( int errnum )
	{
	if ( errnum >= _PDCLIB_ERRNO_MAX || errnum < 0 )
	{
	return _PDCLIB_EUNKNOWN_TEXT;
	}
	else
	{
	return _PDCLIB_lc_messages.errno_texts[errnum];
	}
	}

We actually discussed this briefly downstream in this conversation:

• We have fairly large gaps in our error numbers / strings (so an array is not ideal).
• There is some ambiguity wether the error message is "unknown error" from the host OS, or if it is "unknown error" from the pdclib error_readout tool.
• There is no way to implement custom "unknown error"-messages algorithmically (for including the error number for example).

[DevSolar]

Hmm...

The benefit of having strerror() / perror() work on an array like that is that it allows for static initialization in _PDCLIB_stdinit.c. I was hoping to get around any "library initialization" plumbing, i.e. a function that would have to be called by the C runtime prior to jumping into main().

It also kept the internals of strerror() / perror() comparatively simple.

I have to think a bit about alternatives... perhaps a two-level mapping, with one array mapping the individual (host) error numbers to a consecutive sequence?

Any ideas, or preferences?

[DevSolar]

The source comment about "static array is not the way to go" was aimed at the original construct, which would not have allowed changing the error messages according to locale. This is almost-solved work-in-progress, which however didn't address the issue of enumeration gaps.

I think I got a solution:

• omit "unknown" error IDs from the array;
• instead, store the error ID in an additional field alongside the message;
• implement strerror / perror as a loop through existing entries, checking them for matching error ID or going "unknown error" if no match is found.

I don't quite see the necessity for having the error number included in the messages generated by strerror / perror. The caller has the value of errno, and can add it to the error output if so desired. Or am I missing something?

[JayFoxRox]

I think I got a solution:

Yes, this was mostly was implying, too:

• Have the error readout tool output NULL for unknown errors.
• Turn the readout tool output into a switch/case which returns only known strings with default case returning NULL.
• Handling of actual unknown error messages in platform specific strerror code can be done into a static variable (strerror in my linux man pages allows this, I assume the C-spec allows this, too?).

There'd be no pre-processing (library initialization). And on platforms that want a number as part of the error message that could be added easily without many code-changes.

I don't quite see the necessity for having the error number included in the messages generated by strerror / perror.

I think it's still useful to allow this for mimicking an existing libc (similar to what the error readout tool achieves).

The caller has the value of errno, and can add it to the error output if so desired.

Manually adding it in the caller means a lot of redundant code and added complexity (as the users code suddenly has to deal with sprintf or something).

[DevSolar]

{Removed revision} does introduce a different structure for holding the error messages. There are no longer holes in _PDCLIB_messages_t. This however required to add the errno value to the array (as the index no longer corresponds to that), and adding some logic to find the correct enrty. (I also added a bit extra so that, if more than one symbol is defined to the same value, always the message for the first symbol is returned.)

In the end, this increases code size, by about 300 bytes on my machine for a release build. (And that is without the code that would add the errno value to the "Unknown Error" message.)

At which point I'd be tempted to roll back that revision. Any thoughts?

---

Edit: To avoid confusion, I have rolled back that last SVN revision for now. A patch tarball is available at http://rootdirectory.ddns.net/static/pdclib_path_strerror.tgz if you want to try some things.

I think a patch that actually increases memory usage wouldn't satisfy anyway, correct?

[DevSolar]

...on second thought, I guess I over-engineered that approach a little. I could write a switch in _PDCLIB_geterrtext() that transposes the non-continuous errno values into continuous index values, and stick with the direct index lookup into _PDCLIB_lc_messages...

I will give that a shot, possibly this weekend.