This Week on p5p 1999/10/24



Notes

It is hard to keep track of everything that happens. As before, please let me know if you have any corrections or additions. Send them to mjd-perl-thisweek-YYYYMM@plover.com where YYYYMM is the current year and month.

You can subscribe to an email version of this summary by sending an empty message to p5p-digest-subscribe@plover.com.

$^O

There was a gigantic discussion of $^O and related matters. This was brought on by Tom, who wants to write a program that cross-checks the SEE ALSO sections of the man pages. The problem: Every version of Linux has a man command that is slightly incompatible with every other. In particular, each system has a different idea of where the pages are and how they are organized. Tom wants his program to find out what sort of Linux it is on, `Red Hat' or `Debian' or whatever, but $^O (and also the uname command) only says linux, which is not enough.

Various discussion ensued. Suggestion 1: Make $^O look like linux-redhat or something. Objections: Changing $^O will break stupid programs that have $^O eq 'linux' instead of $^O =~ /linux/. Putting redhat into $^O will not actually solve Tom's problem, at least not in general, since the semantics of redhat changes from release to release.

Suggestion 2: Add a Config.pm field for the distribution vendor. Objections: Config.pm only reflects the state of the system at the time Perl was built, not at the time your program runs. Possible solution to this: Have Config determine the OS at run time at the moment the information is requested. Second objection: If Config can do this, why can't Tom's program do it the same way, but without Config? Well, OK, the nastiness could be encapsulated in a module. But Sarathy didn't like the idea of putting this dynamic information into Config. He suggested:

Suggestion 3: A new module, OS, to provide functions for looking up this sort of thing dynamically. There were other similar suggestions. Dan Sugalski suggested adding a new magical %^O variable that would behave similarly. Nick Ing-Simmons suggested an OS_Info module. This multiplicity suggests that I was the only one following the whole tedious discussion. (And, if so, that everyone else had good sense.)

Gosh. When I took this job, I knew there would be occasional weeks where there was some gigantic but trivial discussion. But I wasn't expecting one so soon.

If there was a conclusion to this discussion, I was not able to find it. Maybe there will be an update next week, or maybe everyone will just get tired of the whole thing and forget about it. Tom eventually punted on the problem, and his program now assumes that it is running under Red Hat.

In this midst of this, there were some sidetracks I found interesting. There was discussion of Sarathy's hack to create fork() on forkless Microsoft OSes (more about this below.) Tom Horsley had a really delightful rant about Configure, which unfortunately is too long to reproduce here:

[Configure] acts, in fact, as though it were a compressed archive chock full of config.h files for all kinds of different systems, and pressing the button merely unpacks one of the files.

The problem comes when you attempt to extract a file that was never put into the archive in the first place. ...

The replies to this are worth reading too.

STOP blocks and the broken compiler

One of the changes in perl 5.005_62 was that END blocks would no longer be run under -c mode. Nick Ing-Simmons wanted to know how the compiler would work; it had formerly worked by enabling -c mode, and walking the op tree and dumping out the compiled code in an END block, which was executed after the program file was parsed and compiled. (This may be an incorrect description; I would be grateful for corrections here.) Disabling END blocks under -c mode, while correct, would break the compiler.

When he made the change, Sarathy planned a workaround, which you can find in perldelta if you are interested. But the workaround is annoying for the compiler, and Sarathy suggested that the best solution would be STOP blocks. These would be run after the compilation phase, but before the run phase; they are in contrast to INIT blocks, which are run at the start of the run phase. Normally, these two things happen at almost the same time, with STOP blocks immediately before INIT blocks. But if you think of a compiler module, which pauses after the compilation phase, writes out the compiled code and exits, the usefulness of STOP becomes clear.

Vishal Bhatia pointed out that this would solve an existing compiler bug: END blocks are presently not executed at all by compiled scripts. If the B:: modules did their work in STOP blocks instead of END blocks, they would not have to usurp the END blocks.

Blank lines in POD

Larry Virden submitted a minor doc patch: There was a line which looked empty, but which contained white space. This prevented the POD parser from recognizing a =head directive on the following line, because directives are only recognized when they begin `paragraphs', and a line is not deemed to end a paragraph unless it is entirely empty.

It appears that this annoying behavior is finally going to be fixed. I am delighted, because I had complained about this back in 1995.

PERL_HEADER environment variable

Ed Peschko wanted a new PERL_HEADER environment variable, somewhat analogous to PERLLIB or PERL5OPT, which would contain code that would be prepended to the source file before it was executed. He wanted this so that he could make an environment setting to tell Perl to always load up some standard, locally defined modules before compiling the rest of any program.

Many people found persuasive reasons why this would be a bad thing to do, and many other people suggested ways that it could be accomplished. For example, you could set PERL5OPT to -MFoo -MBar.

Out of date modules in Perl distribution

Michael Schwern pointed out that there are several modules being distributed with Perl for which more recent versions exist on CPAN.

It turns out that many of these cases are for good reasons. For example, Ilya keeps the version number of the Devel::Peek on CPAN higher than the version in Perl so that if you ask CPAN.pm to install Devel::Peek, it does not go and try to install the latest version of Perl for you. (Why does it do that, anyway?)

However, some modules really are out of date in the distribution. Sarathy asked that authors of modules in the Perl distribution send him a note when they update their modules.

Enhanced UNIVERSAL::isa

Mark Mielke suggested enhancing isa so that you could give it and object and several class names and it would return true if the object belonged to any of the classes. At present, only one class is allowed. No conclusion was reached. My guess is that this is not going in because it is easy to write such a function if you want it.

sort improvements

I don't fully understand this yet, but it looks interesting. It appears that Peter Haworth wants to have Perl notice when a sort comparator function is prototyped with ($$), and to optimize the argument passing to such a function to get the speed of the $a-$b hack, but without actually using $a and $b. Then you could use any two-argument function as a sort comparator but it would be as fast as if it were using the special $a-$b method. I have asked Peter to confirm this, and I will report back next week.

Note added 26 October: Peter cofirms that I have it mostly right, but adds:

The gains aren't so much for performance, as getting rid of package annoyances. If I manage to get this patch working properly, you can use a comparator function from a different package, and it can just get its arguments from @_, rather than ${caller.'::a'} and ${caller.'::b'}. Also, Ilya says this will allow XSUBs to be used as comparators, but I don't know the history of this well enough to know why they can't be used now.

glob case-sensitivity

Perl 5.005_62 optionally has a new built-in implementation of the glob function; it does not need to call the shell to do a glob. Paul Moore pointed out that the new internal globber is case-sensitive, even on his Win32 system with the case-insensitive filesystem; formerly, glob had been case-insensitive.

Some discussion ensued about what to do. Sarathy seemed inclined to let the new globber continue to be insensitive on case-insensitive filesystems, and vice versa; on Windows systems there is an API for finding this out. He asked Paul for a patch for this. He said that people could use the File::Glob or File::DosGlob modules if they needed a specific semantics.

Incidentally, Larry suggested that the new glob be made the default for the beta test versions of Perl, so that it would be tested adequately.

reftype function

Jeff Pinyan posted a complaint about the behavior of a function prototyped with (;$). He wants print f arg1, arg2 to be parsed as if he had written print f(arg1), arg2. At present, Perl aborts, complaining that f got two arguments and expected at most one. Jeff encountered this behavior while he was writing a function to determine what kind of reference (array, hash, whatever) its argument is.

(This is more difficult than it seems. You cannot use only ref, because if you have an object blessed into a class named ARRAY, ref will return ARRAY even if the object is a hash, and you run into similar problems with classes named 0 and so forth.)

Nobody addressed the (;$) issue, but there was discussion of how to build such a function. Spider Boardman revealed that he had such a function named attributes::reftype already in the standard Perl distribution. It is written in C as an XS, which is clearly the Right Way to Do It. Sarathy said he thought that attribute.pm was a good place for the function to be.

New perlthread man page

Dan Sugalski presented for comments a draft of a perlthread man page, discussing Perl's thread interface and thread semantics.

Win32 and fork()

Sarathy has been working for some time on making fork work on forkless Win32 systems. The idea: fork will create a new thread, running a separate copy of the Perl interpreter, which will run the fictional child process. The child process will somehow have its own current working directory, environment, open file table, and so forth. exec in the `child' thread will terminate the thread and its associated interpreter, rather than the entire process.

Dan Sugalski: I see there's going to be something interesting to implement for VMS before 5.6 gets released. Cool. :)

Module Bundling and the proposed import pragma

This continued from last week. Michael King split up his module functionality into Import::ShortName for module aliasing, and Import::JavaPkg, to load a whole bunch of modules in a single namespace all at once, with aliasing.

At the tail end of this discussion, several people complained that although they thought that they'd followed the documented procedure for reserving namespaces in the CPAN module list, nothing ever seemed to come of it, and their names never appeared in the list. Andreas König took responsibility for this problem. He is rewriting the PAUSE software to handle the bookkeeping, because the module list owners are too overworked to do it all manually.

Andreas asked people whose requests had been forgotten to send a reminder to the module list by the end of October, and promised to get these requests listed within 24 hours.

cron daemon runs processes with $SIG{CHLD} set to IGNORE

On some systems, the cron daemon has this bug. (It is a bug in cron, because cron should know to restore the signal handling to the default case when running a job; otherwise the job will inherit this unusual signal environment and might get unexpected results.)

Tom Phoenix added a patch to the linux hints file to try to detect this, and print out a warning at Perl build time if so. Sarathy said it was bad to put this in the hints, because it does not actually affect the build process, and that it should be documented more prominently.

Mike Guy asked: ``Wouldn't it be better for Perl just to set $SIG{CHLD} = 'DEFAULT' automatically at startup in this case? Would it do any harm to do it in all cases?'' Sarathy agreed, and put in a patch to do that, and also to issue a warning if so.

Day range checking in Time::Local::timelocal

If you ask timelocal to convert a date where the day of the month is larger than 31, it aborts with a warning like

        Day '32' out of range 1..31

John L. Allen complained that this was stupid for two reasons: First, it doesn't abort when you ask for February 30, and second, it prevents you from asking for January 280 to find out the date of the 280th day of the year. He submitted a patch that eliminated the check.

A patch like that had been in before, but Sarathy took it out because it caused a test failure in libwww; Sarathy wants it to be conditionalized on a nocroak variable or something, for backward compatibility. In the ensuing discussion, Jonathan Scott Duff made a list of new features he'd like to see in Time::Local---features like `fast' and `correct'.

Mike Guy said that he had worked on such a thing, but run into some annoying backward compatibility issues. For example, the current timelocal returns -1 on an error. But because -1 also indicates a valid time before 1970, timelocal cannot work for dates before 1970 and be backward-compatible with the current version at the same time. Also, the existing timelocal has a very nasty interpretation of the year: 2070, 170, and 70 all mean the year 2070, contrary to good sense and the documentation.

Sarathy said he would accept the timelocal replacement if there were a command to enable the improved behaviors that were not backward compatible with the old behavior.

New quotation characters

Kragen Sitaker asked, on comp.lang.perl.misc, whether it wouldn't be nice for Perl to recognize additional kinds of parentheses once Unicode support is really in. For example, U+3010 and U+3011 are left and right `black lenticular brackets'. The q operator understands q{...} and q(....) q[...] and the like; why not the black lenticular brackets also?

Kragen also suggested that, the Japanese `corner quote' characters U+300C and U+300D (for example) could be used to imply the qr operator, in the same way that ordinary double quotes presently imply the qq operator and ordinary backquotes imply the qx operator.

Ilya thought it was worth forwarding to p5p: ``Once Unicode goes in, one would not be able to change matching rules. So it should be at least discussed early.'' But nobody had anything to say about it.

Lexical or dynamic scope for use utf8?

It is presently lexically scoped. There was discussion some weeks ago about whether to make it dynamically scoped; then the caller of a function could set the utf8 behavior of the library functions it called. I did not understand the issues at the time, so I cannot rehash them here.

Sarathy asked for informed persons to contribute their thoughts, but there were none.

Full path of cwd in @INC

Ed Peschko asked if it would be possible to include the full path of the current directory in @INC, rather than just a dot. The usual objections: 1. There is already an easy way to put the full path in, if that is what you want: you use the FindBin module. 2. It would be expensive for the large population that did not need it.

A Strategic Decision to use the Perl Compiler

Sounds like a bad move to me, but David Falk had this to say for himself:

I am CEO of Dionaea Corporation, a software company that designs performance monitoring tools for UNIX, and we have made the strategic decision to use the perlcc compiler as the hub of our code development. Overall this has been a good decision for us, but we have run into several snags with the compiler.

He then reported the bugs. They looked pretty simple, but nobody replied. Scary to think that someone's family might starve in the streets because of problems in the Perl compiler.

Happy Birthday Perl 5

Actually the real birthday was on 17 October, 1994, but there is an error in perlhist so the birthday wishes arrived on the 18th. (Nobody has supplied a patch yet.)

Chris Nandor submitted a birthday patch.

Unicode Character Classes Revisited

Last week there was discussion of use of Unicode properties to define regex character classes. People interested should also consider reading the Unicode Regular Expression Guidelines.

Sarathy says `Yikes' again

``Yikes, this is one is the size of China.''

Various

A large collection of bug reports, bug fixes, non-bug reports, questions, answers, and a small amount of flamage and spam.

Also, Tuomas Lukka continues to send email with an incorrect Date: header.

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


Mark-Jason Dominus
Visit the home of the Perl programming language: Perl.org

Sponsored by

Monthly Archives

Powered by Movable Type 5.13-en