This Week on p5p 2000/04/30


You can subscribe to an email version of this summary by sending an empty message to

Please send corrections and additions to where YYYYMM is the current year and month.

Line Disciplines Continue

The discussion of Simon’s trial implementation of line disciplines continued, with Sarathy, Ilya, Nick Ing-Simmons, and Larry participating. Some topics of discussion: How do disciplines interact with end-of-line location? Will a discipline request data character-by-character from the discipline below it, or can it request lines or larger chunks? Does a filter transform a file object into another file object, or does it transform an SV into another SV? In either case, what to do with the current sv_gets function that reads a line from a file and passes the resulting line upwards as an SV? What is the real meaning of :raw? Is it a filter, or an indication that filters should not be used? What happens to $/ in the presence of disciplines? What does sysread do when buffering is being provided by some line discipline? (Larry’s answer to this last: It calls the low-level read() function, bypassing the disciplines, as usual.)

Paul Moore asked what the benefit of line disciplines is that is not shared by tied filehandles. The answer was ‘speed’. Tied filehandles have all the overhead of objects and method calls.

Ben Tilly Fixes map

Last week, I reported on all the interesting topics raised by Ben Tilly, including an opservation that the grep code is optimized for the assumption that the output will be smaller than (or the same size as) the input, which is a good assumption. However, the map code, which is derived from grep, is optimized the same way, but in this case the assumption is totally wrong. Ben pointed out that you can rewrite a version of map in Perl that is faster than the builtin map for many common cases. He even suggested how this might be fixed, and I regretfully observed that he did not provide a patch.

Apparently spurred by this, Ben submitted a patch! Woo-hoo!

Gack! Big Bug in 5.6

Larry Rosler pointed out that

        perl -e "my $x = 10; $x = '2' . $x; print $x + 0"

prints 10 instead of 210. Whoops. Simon Cozens provided a patch. 5.6.1 anybody?

More pack Options

John Holdsworth added more features to pack] to better support Sun’s XDR (external data representation) formats, which is something I haven’t looked at since about 1990. His change also supports ? as a length specifier, saying that the length can be found in the data. Ilya found fault with this; he thinks it is not general enough.

Interested parties might like to read more

For me this discussion is reminiscent of a long discussion from last year in which Ilya wanted to introduce a feature that would make data self-unpacking; it would carry its own template with it.

Similar discussion from last year


Barrie Slaymaker posted a really interesting module that combines open3, select and , waitpid. This is something we’ve needed for a long time. Everyone should check out this thread.

Full details, including man page

Discussion about the user interface followed for about 25 messages.

perldiag Re-alphabetized

I submitted a little patch to add a new diagnostic message, and Ron Kimball pointed out that I had put the message into the wrong part of perldiag. The messages in perldiag are supposed to be in alphabetical order, but in 5.6 there are a bunch that are just tacked onto the end in their own miniature alphabetical order, and I had added mine there by mistake.

I re-alphabetized the whole thing with a different (and more consistent) alphabetic scheme.

Novus Ordo Seclorum

Poor Diagnosis of sort Errors

If you try to use a sort comparator function when either of the magical variables $a or $b has been declared to have lexical scope, the comparator won’t work, because it is looking at the lexical variables, and sort is leaving the values to be compared in the global variables. There’s a diagnostic for this:

        Can't use "my $a" in sort comparison

Here’s what perldiag says about it:

     (F) The global variables $a and $b are reserved for sort comparisons.
     You mentioned $a or $b in the same line as the <=> or cmp operator,
     and the variable had earlier been declared as a lexical variable.
     Either qualify the sort variable with the package name, or rename the
     lexical variable.

Peter Scott points out that this is not very nice—it only detects uses of $a that appear in comparisons; in particular, the following code does not produce the diagnostic, although it should:

        my $a;
        my @sorted_list = sort { sortfunc($a, $b) } @list;

I know there are many people out there looking for a not-too-difficult core hacking project to get started on. Fixing this might be such a project.

glob() function not documented

Randal pointed out that the File::Glob man page, which is the default globber for the 5.6.0 glob() function, does not say what the function actually does. For example, does this globber support the csh-style {foo,bar,baz} notation? Perhaps someone would like to rectify this?

Overloading Assignment

You can’t use the overload mechanism to overload assignment, but you can use the tie mechanism. Jeff Pinyan posted an example module that wraps up overloaded assignment semantics for a module. Damian replied that he has a paper about that. Discussion went offlist.

Jeff’s Idea

POSIX Character Classes

POSIX Character Classes (new in 5.6) do not work in the tr operator. Since \d and \w, and character classes generally also do not work there, it is not a big surprise. Summary: tr[a-z][A-Z] looks like it has character classes, but it doesn’t.

First Cite for ‘Perl’?

The Oxford English Dictionary folks contacted Larry and asked for the earliest possible print citation for the word ‘Perl’. (The OED is the premier dictionary of the English languagem, and seeks to provide contemporary quotations for every word; that way you can see from examples how the meaning and usage of each word has changed over its lifetime. The Third Edition of the OED is due out in 2007.)

The earliest cite located seems to be around 1989; it would be good to get an earlier cite, which the OED folks call an ‘antedating’. If you know of an English source that mentions Perl that was printed prior to 1989, please contact Larry. If you can’t reach Larry, contact me.

People guess that the best bet for antedatings is probably old ;login: magazines. If you have a pile of these, please take a look.


A large collection of bug reports, bug fixes, an unusual number of non-bug reports (split deletes trailing null fields! int rounds down! $object->$field = $value is a snytax error!), questions, answers, and a small amaount of spam. There were no flames this week.

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

Mark-Jason Dominus



Something wrong with this article? Help us out by opening an issue or pull request on GitHub