Re^5: Determining the minimum representable increment/decrement possible? (Perl's scanf broken?)

[syphilis]

Why not use the underlying CRTs sscanf() or strtod()?

I don't know.
I speculate that in earlier times those functions were not so reliable and it was decided that perl should do its own implementation.
I further speculate that it has now got to the stage that unwinding that perl implementation is not a trivial task.

Cheers,
Rob

[BrowserUk]

Grr. I had a notion to try and work out what code Perl was actually using to parse doubles and 'forgot' my rule of the last few years: don't go looking in the recent perl sources; it will only piss you off.

It pissed me off. It used to be that you could compile Perl to use stdio, and I thought I might do that for 5.22; but apparently that is not longer possible!

Perl is still my go-to language for getting stuff done in a hurry; and in many cases even when the program has ultimately to be written in some compiled language, whether for speed or to satisfy some other requirement, I still use Perl to prototype and only convert to C or C++ or D or whatever once I have the code working.

But for the last few years, the P5 development has become introspective, even introverted; titivating instead of innovating; throwing any weird idea that comes up into the code base only to remove it a build or two later. I've lost track of the number of "experimental" additions that have come and inevitably gone.

It's just a shame that PerlIO isn't so easily back out and discarded! :(

[syphilis]

I hadn't thought of building perl without USE_PERLIO.
That would have been interesting - and I agree that it's a pity that the option has been taken away.

Cheers,
Rob

[syphilis]

I hadn't thought of building perl without USE_PERLIO

BTW, in a recent email I suggested to Jarkko Hietaniemi that building without USE_PERLIO might be helpful re perl's problems with text-to-float.

He said "I don't think the float conversions have anything to do with perlio as such. Look at numeric.c."

Cheers,
Rob

[ikegami]

Only two experimental features were removed:

• Passing a reference to key/values/each

  I don't know why that was released in the first place given it was known to be broken before it was even coded.

• my $_;

  It was causing code like apply { s/foo/bar/ } @a to break in a manner that was confusing for too many.

Smart-matching and given/when have been experimental for a while, but they are not dead. The subject came up again this week. There are strong intentions to fix it.