This Week on p5p 2001/03/12

Mar 12, 2001 by Simon Cozens

Notes

You can subscribe to an email version of this summary by sending an empty message to perl5-porters-digest-subscribe@netthink.co.uk.

Please send corrections and additions to perl-thisweek-YYYYMM@simon-cozens.org where YYYYMM is the current year and month. Changes and additions to the perl5-porters biographies are particularly welcome.

There were 424 messages this week.

Pod Questions

As reported last week, Michael Stevens has been working away on attempting to make the core Perl documentation podchecker-clean, and has succeeded in stopping it from emitting any errors. However, he came up with quite a few weirdnesses. The most contentious was the correct way to write:

     L<New C<qr//> operator>

since L<> was seeing the slash and thinking it was a section/manpage separator. Russ Allbery said that the best way was

    L<"New C<qr//> operator">

but the problem with that is that the resulting reference gets quoted. And, in fact, podchecker was still unhappy with that. Russ said:

podchecker complains about all sorts of things that I consider to be perfectly valid POD, such as the use of < and > in free text to mean themselves when not preceeded by a capital letter. I think making podchecker smarter is the right solution.

But as Michael said, “the problem is finding a clear definition of what “smarter” actually is.”

I also complained that

    =head2 New C<qr//> operator

was getting mangled by some parsers which didn’t correctly restore boldface after the code section. The example I gave, pod2man, seemed to be due to a buggy set of roff macros.

Rob Napier came up with some truly excellent suggestions about the future of POD and how to make it more intuitive, and Russ tried to shoo people onto the pod-people mailing list for further discussion of what changes should be made.

Patching perly.y

Jeff Pinyan asked how one should go about patching the Perl grammar in perly.y; the answer, coming in three parts from myself, Peter Prymmer and Dan Sugalski, is:

1) Don’t. You hardly ever need to.

2) Run make run_byacc which runs the byacc parser generator, and then applies a small patch to the resulting C file which allows dynamic memory allocation.

3) Run vms/vms_yfix.pl to patch up the VMS version of the parser.

4) CC perl-mvs@perl.org so that the EBCDIC people can prepare EBCDIC-aware versions of the parser.

CvOUTSIDE

Alan asked what CvOUTSIDE was for; it’s another undocumented flag on a CV. Sarathy knows the answer, and it’s scary:

Every CV that references lexicals from its outer lexical scopes needs to be able to access that outer scope’s scratchpad at run time (via pp_anonsub(), cv_clone2() and pad_findlex()) to capture the lexicals that are visible at the time the cloning happens. In fact, all CVs need to have this whether they have outer lexicals referenced in them or not, given that eval"" requires visibility of the outer lexical scopes.

Hence, (I think) CvOUTSIDE is a pointer to the scratchpad of the outer lexical scope. Why is this important? Well, Alan’s Great Subroutine Memory Leak (the problem with sub x { sub {} }) has come about because there’s a reference count loop. As Sarathy explains:

The problem really is that there is a reference loop. The prototype anonymous sub holds a reference count on the outer sub via CvOUTSIDE(). The outer sub holds a reference count on the anonymous sub prototype via the pad entry allocated by OP_ANONCODE. The pad entry will be properly freed by op_clear() if it ever gets there, which it doesn’t because of the loop.

Sarathy had a couple of attempts at fixing this, but hasn’t managed to resolve it yet.

perlxstut Documentation

Vinh Lam reminded us that perlxstut is incomplete. Examples 6, 7, 8, and 9 are still not written. Does anyone out there want to write them?

EBCDIC and Unicode

With the assistance of Merijn Broeren and Morgan Stanley Dean Witter, I gained access to an EBCDIC mainframe and spent a happy day sanitizing the Unicode support on EBCDIC machines. As usual, there was some small argument over semantics, but the major change was that EBCDIC should be converted to ASCII before being upgraded to UTF8, and converted back to EBCDIC on degradation. Peter Prymmer seemed happy enough with what we’d been doing, and the patch went in. The patch, and its discussion, can be found here.

If you don’t want to read the whole business, this is the important bit: much of the Unicode discussion this week centered on the vexed question of “What are v-strings for?”. Here is the definitive answer from Larry.

PERL_DL_NONLAZY

Michael Schwern asked what the mysterious PERL_DL_NONLAZY environment variable was for - it’s set on make test but never documented. He noted that as well as being used to alter the dynamic linking behaviour, it’s used by some test suites to determine whether or not to produce additional information - almost certainly a misuse.

Paul Johnson explained that it passes a flag to dlopen which attempts to ensure that all functions are relocated as soon as the shared object is loaded. Sounds complicated? In the normal, “lazy” operation of the dynamic loader, the loader doesn’t actually load all the functions from the library file into memory at one go - instead, it merely notices that it has a bunch more functions available; when a function is called, it loads up the appropriate part of the object into memory, and jumps to it. (Not entirely unlike the behaviour of use autouse or AutoSplit.)

Setting [PERL_DL_NONLAZY] forces the loader to load up all functions at once, so that it can ensure that it really does have code for all the functions it claims to have code for; this is usually what you want to do when testing.

Various

Sarathy fixed the “weird reset bug” of last week with a clever but untested patch; Chris Nandor dropped a bunch of good MacPerl protability patches. Ilya finally produced his rival UTF8 regular expressions patch, which Jarkko has been vigorously testing.

David Madison raised the my $var if $x bugbear again. Schwern’s been cleaning up Test::Harness; good work as always, there. Robin Houston fixed a strange bug regarding my variables being cleared after a goto during a three-fingered for loop. Radu Greab fixed something strange with chop and arrays.

There was a small but pointless discussion of C coding styles, which concluded that you ought to leave off braces around single-statement blocks to if and the like if you can.

Tony Finch complained that use integer doesn’t make rand return integers; Philip Newton provided a patch.

Congratulations to Raphael Manfredi, who spawned his first child process this week.

Until next week I remain, your humble and obedient servant,

Simon Cozens - Notes - Pod Questions - Patching perly.y - CvOUTSIDE - perlxstut Documentation - EBCDIC and Unicode - PERL_DL_NONLAZY - Various