Re: Dependency Hell (Test::Harness)
David R. Morrison wrote: I think the problem here is that "perl Makefile.PL" will _use_ the value of PERL5LIB which you have set in order to locate any modules necessary during the creation of the Makefile, but will not _pass_ this value to the Makefile itself in any useful manner. Yes, I think that's exactly it -- and even worse, not only does $PERL5LIB/Test/Harness.pm not get passed, it even gets *overriden* by /System/Library/Perl/Test/Harness.pm, because the Makefile sets PERL_LIB = /System/Library/Perl and then includes it: test_dynamic :: pure_all PERL_DL_NONLAZY=1 $(FULLPERL) -I$(INST_ARCHLIB) -I$(INST_LIB) -I$(PERL_ARCHLIB) -I$(PERL_LIB) -e 'use Test::Harness qw(&runtests $$verbose); $$verbose=$(TEST_VERBOSE); runtests @ARGV;' $(TEST_FILES) (This is written by the "test_via_harness" method of ExtUtils::MM_Unix.) To reverse this, Ken Williams suggested calling "make test" with PERL_LIB=$PERL5LIB, but I'm not yet sure that this is always foolprof. In Fink, "perl Makefile.PL" is called with a bunch of arguments: perl Makefile.PL PREFIX=/sw \ INSTALLPRIVLIB=/sw/lib/perl5 \ INSTALLARCHLIB=/sw/lib/perl5/darwin \ INSTALLSITELIB=/sw/lib/perl5 \ INSTALLSITEARCH=/sw/lib/perl5/darwin \ INSTALLMAN1DIR=/sw/share/man/man1 \ INSTALLMAN3DIR=/sw/share/man/man3 (see the description of CompileScript at http://fink.sourceforge.net/doc/packaging/reference.php#fields ). ...indeed, that's exactly where I took the flags mentioned in the original post :-) (And I'll take this opportunity to thank you for all your Fink work. Maybe some day you'll tell the rest of us how you manage this *and* your day job...) I also browsed Fink CVS to see how you go about my problem for the modules that Fink distributes. (I guess they all come as dependencies for other packages, or is there any other reason for Fink to distribute perl modules?) However it looks like none of them need the new Test::Harness at present... but I guess this is bound to happen sooner or later. How would you go about it? Write a custom "make test"? The relevant variables for you would presumably be the ...LIB ones. Well, looking at the above excerpt from the Makefile, it seems that again, PERL_LIB is the only one we could usefully set (to $PERL5LIB). But I'm afraid that doing this at "perl Makefile.PL" time is too early in general... For this particular module the resulting Makefile happens to work, but the warning (below) leads me to believe that it wouldn't in general, no? (Cf. the problems when I set PERL_ARCHLIB earlier in this thread.) [localhost:~/Getopt-ArgvFile-1.06] fz% perl Makefile.PL PERL_LIB=$PERL5LIB Checking if your kit is complete... Looks good Warning: PERL_LIB (/Volumes/Unix/usrlocal/lib/perl5) seems not to be a perl library directory (Exporter.pm not found) at (eval 9) line 367. Writing Makefile for Getopt::ArgvFile Looking again at the Makefile excerpt, another possibility that comes to mind is to set FULLPERL to "/usr/bin/perl -I$PERL5LIB". This ensures that the newly installed Test::Harness is selected, and it seems to work (see below). However I'm not sure what would happen if we have a module in PERL5LIB and try to install a newer version of it: will "make test" actually test the new one, or the old one? Seems like it could be the latter, and then... does anyone see if/how we could override that? [localhost:~/Getopt-ArgvFile-1.06] fz% perl Makefile.PL FULLPERL="/usr/bin/perl -I$PERL5LIB" Checking if your kit is complete... Looks good Writing Makefile for Getopt::ArgvFile [localhost:~/Getopt-ArgvFile-1.06] fz% make mkdir blib mkdir blib/lib mkdir blib/lib/Getopt mkdir blib/arch mkdir blib/arch/auto mkdir blib/arch/auto/Getopt mkdir blib/arch/auto/Getopt/ArgvFile mkdir blib/lib/auto mkdir blib/lib/auto/Getopt mkdir blib/lib/auto/Getopt/ArgvFile mkdir blib/man3 cp ArgvFile.pm blib/lib/Getopt/ArgvFile.pm Manifying blib/man3/Getopt::ArgvFile.3 [localhost:~/Getopt-ArgvFile-1.06] fz% make test PERL_DL_NONLAZY=1 /usr/bin/perl -I/Volumes/Unix/usrlocal/lib/perl5 -Iblib/arch -Iblib/lib -I/System/Library/Perl/darwin -I/System/Library/Perl -e 'use Test::Harness qw(&runtests $verbose); $verbose=0; runtests @ARGV;' t/*.t t/base..ok t/prefixok All tests successful. Files=2, Tests=7, 1 wallclock secs ( 0.80 cusr + 0.32 csys = 1.12 CPU)
Re: Dependency Hell (Test::Harness)
Ken Williams wrote: So, is there a workaround to force the use of the newer Test::Harness in "make test"? That's a pain. You could probably do this: make test PERL_ARCHLIB=/foo/darwin PERL_LIB=/foo This doesn't quite work for me: [localhost:~/Getopt-ArgvFile-1.06] fz% make test PERL_ARCHLIB=$PERL5LIB/darwin PERL_LIB=$PERL5LIB make: *** No rule to make target `/Volumes/Unix/usrlocal/lib/perl5/darwin/Config.pm', needed by `Makefile'. Stop. but it *does* work (thanks!) if I only set PERL_LIB: [localhost:~/Getopt-ArgvFile-1.06] fz% make test PERL_LIB=$PERL5LIB PERL_DL_NONLAZY=1 /usr/bin/perl -Iblib/arch -Iblib/lib -I/System/Library/Perl/darwin -I/Volumes/Unix/usrlocal/lib/perl5 -e 'use Test::Harness qw(&runtests $verbose); $verbose=0; runtests @ARGV;' t/*.t t/base..ok t/prefixok All tests successful. Files=2, Tests=7, 2 wallclock secs ( 0.81 cusr + 0.41 csys = 1.22 CPU) (It even works with PERL_LIB=/dev/null, as long as I keep PERL5LIB in my environment. PERL_ARCHLIB, on the other hand, can't be changed because at some point the Makefile wants to set CONFIGDEP = $(PERL_ARCHLIB)/Config.pm (etc.)) However, this leaves me with two questions. First, can I hope this to always work? Or was it by special luck in this case that the tests didn't require anything from /System/Library/Perl (which is what the Makefile had PERL_LIB set to)? It looks like what we'd really need here is the ability to specify a library *path*, rather than just one directory... Second, of course, I'd like ultimately to have CPAN do all this for me. Now I see that it lets me set "makepl_arg", "make_arg", "make_install_arg", but nothing like a "make_test_arg" :-( Is there a way? If I were you, though, I'd just install Test::Harness into the standard /Library/Perl/ (Apple knows this is the location users will be installing modules into, that's what that directory is for), and install with sudo make install UNINST=1 which will remove the ones in /System/Library/Perl/ . That sounds like the voice of reason, to which I'll probably surrender :-) Still, if there is a way to segregate our stuff from Apple's -- to override it, rather than replace it -- I think it may be worth finding. After all, shouldn't non-sudoers be able to install modules that need the new Test::Harness? Also it would simplify backups, provide a single switch to revert to the stock behavior (unsetenv PERL5LIB), and insure us against Apple or CPAN changing their minds about who owns /Library/Perl, /System/Library/Perl, or /usr/bin/head... (It even works with PERL_LIB=/dev/null, as long as I keep PERL5LIB in my environment. Apparently the reason we can't also change PERL_ARCHLIB is that the Makefile uses it to set CONFIGDEP.) # Where is the Config information that we are using/depend on CONFIGDEP = $(PERL_ARCHLIB)/Config.pm $(PERL_INC)/config.h )
Re: Dependency Hell (Test::Harness)
On Sunday, February 23, 2003, at 05:06 PM, David Wheeler wrote: On Sunday, February 23, 2003, at 11:07 AM, Sherm Pendley wrote: I recently upgraded to the latest CPAN.pm, and I noticed a difference in behavior - when updating a module that's part of the core, it updates the copy in /System/Library/Perl instead of installing a new copy in /Library/Perl. I noticed that CPANPLUS did this for me, too, with Perl 5.8.0 in /usr/local. It installed Attribute::Handlers in /usr/local/lib/perl5/site_perl/5.8.0 but left the copy in /usr/local/lib/perl5/5.8.0 (core). This leads me to think that this is typical and expected behavior. Yes - what you're describing is the traditional behavior. As Ken pointed out, you needed to use the UNINST=1 switch to delete the old core module. That's why I was surprised, and why I felt it worth commenting on it, when my install of Test::Harness didn't behave that way, when I installed it with the latest CPAN. Instead of installing it in /Library/Perl - the moral equivalent of site_perl, for Apple's built-in 5.6 - it updated the core copy of the module in /System/Library/Perl. I just updated CPAN.pm using 5.8.0 in /opt, and CGI.pm, using both 5.6.0 and 5.8.0, with the same results. Each time, it updated the core module in-place, instead of installing the new version in site_perl as I'd expected it to. sherm-- "I have no special gift, I am only passionately curious." - Albert Einstein
Re: Dependency Hell (Test::Harness)
On Sunday, February 23, 2003, at 11:07 AM, Sherm Pendley wrote: I recently upgraded to the latest CPAN.pm, and I noticed a difference in behavior - when updating a module that's part of the core, it updates the copy in /System/Library/Perl instead of installing a new copy in /Library/Perl. I noticed that CPANPLUS did this for me, too, with Perl 5.8.0 in /usr/local. It installed Attribute::Handlers in /usr/local/lib/perl5/site_perl/5.8.0 but left the copy in /usr/local/lib/perl5/5.8.0 (core). This leads me to think that this is typical and expected behavior. David -- David Wheeler AIM: dwTheory [EMAIL PROTECTED] ICQ: 15726394 Yahoo!: dew7e Jabber: [EMAIL PROTECTED] Kineticode. Setting knowledge in motion.[sm]
Re: Dependency Hell (Test::Harness)
On Sunday, February 23, 2003, at 10:27 AM, Ken Williams wrote: If I were you, though, I'd just install Test::Harness into the standard /Library/Perl/ (Apple knows this is the location users will be installing modules into, that's what that directory is for), and install with sudo make install UNINST=1 which will remove the ones in /System/Library/Perl/ . I recently upgraded to the latest CPAN.pm, and I noticed a difference in behavior - when updating a module that's part of the core, it updates the copy in /System/Library/Perl instead of installing a new copy in /Library/Perl. sherm-- C programmers never die - they're just cast into void.
Re: Dependency Hell (Test::Harness)
On Saturday, February 22, 2003, at 09:14 PM, [EMAIL PROTECTED] wrote: Thanks again. (I already replied once to say that I'm positive PERL5LIB was set, but somehow this hasn't made it to the list yet.) The problem, it seems, is that the "-I/System/Library/Perl" overrides it :-( Viz. the difference if I run the tests manually with and without this flag: [localhost:~/Getopt-ArgvFile-1.06] fz% echo $PERL5LIB /Volumes/Unix/usrlocal/lib/perl5 [localhost:~/Getopt-ArgvFile-1.06] fz% perl -Iblib/arch -Iblib/lib -I/System/Library/Perl/darwin -I/System/Library/Perl -e 'use Test::Harness qw(&runtests $verbose); $verbose=0; runtests @ARGV;' t/*.t t/base..FAILED tests 1-5 Failed 5/5 tests, 0.00% okay t/prefixFAILED tests 1-2 Failed 2/2 tests, 0.00% okay Failed Test Status Wstat Total Fail Failed List of failed --- t/base.t 55 100.00% 1-5 t/prefix.t22 100.00% 1-2 Failed 2/2 test scripts, 0.00% okay. 7/7 subtests failed, 0.00% okay. [localhost:~/Getopt-ArgvFile-1.06] fz% echo $PERL5LIB /Volumes/Unix/usrlocal/lib/perl5 Files=2, Tests=7, 2 wallclock secs ( 0.85 cusr + 0.31 csys = 1.16 CPU) [localhost:~/Getopt-ArgvFile-1.06] fz% perl -Iblib/arch -Iblib/lib -e 'use Test::Harness qw(&runtests $verbose); $verbose=0; runtests @ARGV;' t/*.t t/base..ok t/prefixok All tests successful. Files=2, Tests=7, 1 wallclock secs ( 0.86 cusr + 0.37 csys = 1.23 CPU) So, is there a workaround to force the use of the newer Test::Harness in "make test"? That's a pain. You could probably do this: make test PERL_ARCHLIB=/foo/darwin PERL_LIB=/foo If I were you, though, I'd just install Test::Harness into the standard /Library/Perl/ (Apple knows this is the location users will be installing modules into, that's what that directory is for), and install with sudo make install UNINST=1 which will remove the ones in /System/Library/Perl/ . -Ken
Re: Dependency Hell (Test::Harness)
I think the problem here is that "perl Makefile.PL" will _use_ the value of PERL5LIB which you have set in order to locate any modules necessary during the creation of the Makefile, but will not _pass_ this value to the Makefile itself in any useful manner. In Fink, "perl Makefile.PL" is called with a bunch of arguments: perl Makefile.PL PREFIX=/sw \ INSTALLPRIVLIB=/sw/lib/perl5 \ INSTALLARCHLIB=/sw/lib/perl5/darwin \ INSTALLSITELIB=/sw/lib/perl5 \ INSTALLSITEARCH=/sw/lib/perl5/darwin \ INSTALLMAN1DIR=/sw/share/man/man1 \ INSTALLMAN3DIR=/sw/share/man/man3 (see the description of CompileScript at http://fink.sourceforge.net/doc/packaging/reference.php#fields ). The relevant variables for you would presumably be the ...LIB ones. -- Dave
Re: Dependency Hell (Test::Harness)
Ken Williams wrote: Looks like maybe PERL5LIB isn't set anymore? It always need to be set in order to find stuff in your non-standard location. Thanks. I think it's still set... Let me start over and show you the exact command sequence: [localhost:~] fz% echo $PERL5LIB /Volumes/Unix/usrlocal/lib/perl5 [localhost:~] fz% tar xzf Getopt-ArgvFile-1.06.tgz [localhost:~] fz% cd Getopt-ArgvFile-1.06 [localhost:~/Getopt-ArgvFile-1.06] fz% perl Makefile.PL `cat ~/.perl_inst_dirs` Checking if your kit is complete... Looks good Writing Makefile for Getopt::ArgvFile [localhost:~/Getopt-ArgvFile-1.06] fz% make mkdir blib mkdir blib/lib mkdir blib/lib/Getopt mkdir blib/arch mkdir blib/arch/auto mkdir blib/arch/auto/Getopt mkdir blib/arch/auto/Getopt/ArgvFile mkdir blib/lib/auto mkdir blib/lib/auto/Getopt mkdir blib/lib/auto/Getopt/ArgvFile mkdir blib/man3 cp ArgvFile.pm blib/lib/Getopt/ArgvFile.pm Manifying blib/man3/Getopt::ArgvFile.3 [localhost:~/Getopt-ArgvFile-1.06] fz% make test PERL_DL_NONLAZY=1 /usr/bin/perl -Iblib/arch -Iblib/lib -I/System/Library/Perl/darwin -I/System/Library/Perl -e 'use Test::Harness qw(&runtests $verbose); $verbose=0; runtests @ARGV;' t/*.t t/base..FAILED tests 1-5 Failed 5/5 tests, 0.00% okay t/prefixFAILED tests 1-2 Failed 2/2 tests, 0.00% okay Failed Test Status Wstat Total Fail Failed List of failed --- t/base.t 55 100.00% 1-5 t/prefix.t22 100.00% 1-2 Failed 2/2 test scripts, 0.00% okay. 7/7 subtests failed, 0.00% okay. make: *** [test_dynamic] Error 255 [localhost:~/Getopt-ArgvFile-1.06] fz% echo $PERL5LIB /Volumes/Unix/usrlocal/lib/perl5 [localhost:~/Getopt-ArgvFile-1.06] fz% perl -V Characteristics of this binary (from libperl): Compile-time options: USE_LARGE_FILES Built under darwin Compiled at Jul 14 2002 04:04:33 %ENV: PERL5LIB="/Volumes/Unix/usrlocal/lib/perl5" @INC: /Volumes/Unix/usrlocal/lib/perl5/darwin /Volumes/Unix/usrlocal/lib/perl5 /System/Library/Perl/darwin /System/Library/Perl /Library/Perl/darwin /Library/Perl /Library/Perl /Network/Library/Perl/darwin /Network/Library/Perl /Network/Library/Perl . On the other hand, in the meantime I had no trouble installing HTML::TextToHTML (which was my target and SAYS it requires Test::More), plus its dependencies ExtUtils::configPL and Filter. The tests ran fine, though again I'm not sure *which* Test::Harness was being used: [localhost:~/Filter-1.29] fz% make test PERL_DL_NONLAZY=1 /usr/bin/perl -Iblib/arch -Iblib/lib -I/System/Library/Perl/darwin -I/System/Library/Perl -e 'use Test::Harness qw(&runtests $verbose); $verbose=0; runtests @ARGV;' t/*.t t/call..ok t/cpp...ok t/decrypt...ok t/exec..ok t/order.ok t/shok t/tee...ok All tests successful. Files=7, Tests=48, 9 wallclock secs ( 4.35 cusr + 2.30 csys = 6.65 CPU) No tests defined for Filter::Util::Call extension. No tests defined for Filter::Util::Exec extension. No tests defined for Filter::decrypt extension. No tests defined for Filter::tee extension. [localhost:~/ExtUtils-configPL-1.1] fz% make test PERL_DL_NONLAZY=1 /usr/bin/perl -Iblib/arch -Iblib/lib -I/System/Library/Perl/darwin -I/System/Library/Perl -e 'use Test::Harness qw(&runtests $verbose); $verbose=0; runtests @ARGV;' t/*.t t/def...ok t/mode..ok t/nook t/opt...ok All tests successful. Files=4, Tests=10, 1 wallclock secs ( 0.17 cusr + 0.05 csys = 0.22 CPU) [localhost:~/HTML-TextToHTML-1.12] fz% make test PERL_DL_NONLAZY=1 /usr/bin/perl -Iblib/arch -Iblib/lib -I/System/Library/Perl/darwin -I/System/Library/Perl test.pl 1..11 ok 1 ok 2 - new() returned something ok 3 - and it's the right class ok 4 - converted sample.txt ok 5 - test file matches original example exactly ok 6 - converted sample string ok 7 - compare converted string with OK string ok 8 - converted sample string with list ok 9 - compare converted list string with OK list string ok 10 - converted xhtml sample.txt ok 11 - test file xhtml_sample.html matches original good_xhtml_sample.html exactly
Re: Dependency Hell (Test::Harness)
Ken Williams wrote: On Saturday, February 22, 2003, at 12:19 PM, [EMAIL PROTECTED] wrote: [localhost:~/Getopt-ArgvFile-1.06] fz% make test PERL_DL_NONLAZY=1 /usr/bin/perl -Iblib/arch -Iblib/lib -I/System/Library/Perl/darwin -I/System/Library/Perl -e 'use Test::Harness qw(&runtests $verbose); $verbose=0; runtests @ARGV;' t/*.t t/base..FAILED tests 1-5 Failed 5/5 tests, 0.00% okay t/prefixFAILED tests 1-2 Failed 2/2 tests, 0.00% okay Failed Test Status Wstat Total Fail Failed List of failed --- t/base.t 55 100.00% 1-5 t/prefix.t22 100.00% 1-2 Failed 2/2 test scripts, 0.00% okay. 7/7 subtests failed, 0.00% okay. make: *** [test_dynamic] Error 255 So, is it at this point using the correct Test::Harness? If yes, what am I doing wrong? If no, what am I doing wrong? Looks like maybe PERL5LIB isn't set anymore? It always need to be set in order to find stuff in your non-standard location. Thanks again. (I already replied once to say that I'm positive PERL5LIB was set, but somehow this hasn't made it to the list yet.) The problem, it seems, is that the "-I/System/Library/Perl" overrides it :-( Viz. the difference if I run the tests manually with and without this flag: [localhost:~/Getopt-ArgvFile-1.06] fz% echo $PERL5LIB /Volumes/Unix/usrlocal/lib/perl5 [localhost:~/Getopt-ArgvFile-1.06] fz% perl -Iblib/arch -Iblib/lib -I/System/Library/Perl/darwin -I/System/Library/Perl -e 'use Test::Harness qw(&runtests $verbose); $verbose=0; runtests @ARGV;' t/*.t t/base..FAILED tests 1-5 Failed 5/5 tests, 0.00% okay t/prefixFAILED tests 1-2 Failed 2/2 tests, 0.00% okay Failed Test Status Wstat Total Fail Failed List of failed --- t/base.t 55 100.00% 1-5 t/prefix.t22 100.00% 1-2 Failed 2/2 test scripts, 0.00% okay. 7/7 subtests failed, 0.00% okay. [localhost:~/Getopt-ArgvFile-1.06] fz% echo $PERL5LIB /Volumes/Unix/usrlocal/lib/perl5 Files=2, Tests=7, 2 wallclock secs ( 0.85 cusr + 0.31 csys = 1.16 CPU) [localhost:~/Getopt-ArgvFile-1.06] fz% perl -Iblib/arch -Iblib/lib -e 'use Test::Harness qw(&runtests $verbose); $verbose=0; runtests @ARGV;' t/*.t t/base..ok t/prefixok All tests successful. Files=2, Tests=7, 1 wallclock secs ( 0.86 cusr + 0.37 csys = 1.23 CPU) So, is there a workaround to force the use of the newer Test::Harness in "make test"?
Re: Dependency Hell (Test::Harness)
On Saturday, February 22, 2003, at 12:19 PM, [EMAIL PROTECTED] wrote: [localhost:~/Getopt-ArgvFile-1.06] fz% make test PERL_DL_NONLAZY=1 /usr/bin/perl -Iblib/arch -Iblib/lib -I/System/Library/Perl/darwin -I/System/Library/Perl -e 'use Test::Harness qw(&runtests $verbose); $verbose=0; runtests @ARGV;' t/*.t t/base..FAILED tests 1-5 Failed 5/5 tests, 0.00% okay t/prefixFAILED tests 1-2 Failed 2/2 tests, 0.00% okay Failed Test Status Wstat Total Fail Failed List of failed --- t/base.t 55 100.00% 1-5 t/prefix.t22 100.00% 1-2 Failed 2/2 test scripts, 0.00% okay. 7/7 subtests failed, 0.00% okay. make: *** [test_dynamic] Error 255 So, is it at this point using the correct Test::Harness? If yes, what am I doing wrong? If no, what am I doing wrong? Looks like maybe PERL5LIB isn't set anymore? It always need to be set in order to find stuff in your non-standard location. -Ken
