ExtUtils-ParseXS lacks test coverage of edge cases #23859
Description
I've been studying (better put: struggling with) #23846 for the past few days. I believe that CPAN distro Authen-SASL-XS has been unmaintained for so long that it may be unsalvageable. (At the very least, its needs a new maintainer). But what I want to discuss in this ticket is what the test failures discussed in that ticket show us about limitations in the coverage which ExtUtils-ParseXS's test suite provides to that distribution's code.
I was not able to get Authen-SASL-XS's Makefile.PL to complete on Debian Linux; I could not figure out what packages to install to get sasl.h. Hence, all my data is drawn from my attempts at running that distro through CPANtesters on FreeBSD-14, where I had the necessary C header files already installed. You can view my CPANtesters reports here.
What I am specifically concerned with is my attempt to build that distro against v5.43.3; that is the subject of this report. Note this output from make:
"/usr/home/jkeenan/testing/perl-5.43.3/bin/perl" "/home/jkeenan/testing/perl-5.43.3/lib/5.43.3/ExtUtils/xsubpp" -typemap '/home/jkeenan/testing/perl-5.43.3/lib/5.43.3/ExtUtils/typemap' -typemap '/usr/home/jkeenan/.cpan/build/Authen-SASL-XS-1.00-1/typemap' XS.xs > XS.xsc
Use of uninitialized value in scalar chomp at /home/jkeenan/testing/perl-5.43.3/lib/5.43.3/ExtUtils/ParseXS.pm line 1287, <__ANONIO__> line 2082.
Use of uninitialized value in substitution (s///) at /home/jkeenan/testing/perl-5.43.3/lib/5.43.3/ExtUtils/ParseXS.pm line 1288, <__ANONIO__> line 2082.
Use of uninitialized value in pattern match (m//) at /home/jkeenan/testing/perl-5.43.3/lib/5.43.3/ExtUtils/ParseXS.pm line 1288, <__ANONIO__> line 2082.
Use of uninitialized value in pattern match (m//) at /home/jkeenan/testing/perl-5.43.3/lib/5.43.3/ExtUtils/ParseXS.pm line 1301, <__ANONIO__> line 2082.
Use of uninitialized value in pattern match (m//) at /home/jkeenan/testing/perl-5.43.3/lib/5.43.3/ExtUtils/ParseXS.pm line 1450, <__ANONIO__> line 2082.
Use of uninitialized value in pattern match (m//) at /home/jkeenan/testing/perl-5.43.3/lib/5.43.3/ExtUtils/ParseXS.pm line 1472, <__ANONIO__> line 2082.
Use of uninitialized value in pattern match (m//) at /home/jkeenan/testing/perl-5.43.3/lib/5.43.3/ExtUtils/ParseXS.pm line 1475, <__ANONIO__> line 2082.
Use of uninitialized value in string eq at /home/jkeenan/testing/perl-5.43.3/lib/5.43.3/ExtUtils/ParseXS.pm line 1523, <__ANONIO__> line 2082.
Please specify prototyping behavior for XS.xs (see perlxs manual)
mv XS.xsc XS.c
These warnings are emitted before the problems reported by @Leont on Oct 15 in #23846 (comment); I believe they are logically distinct from those problems. What concerns me is that ExtUtils-ParseXS's test suite is apparently not exercising ("covering") the statements in ParseXS.pm where those Use of uninitialized value in ... warnings are being emitted. Here are the 3 relevant subroutines with comments indicating the warnings lines.
1278 sub _maybe_skip_pod {
1279 my ExtUtils::ParseXS $self = shift;
1280
1281 while ($self->{lastline} =~ /^=/) {
1282 while ($self->{lastline} = readline($self->{in_fh})) {
1283 last if ($self->{lastline} =~ /^=cut\s*$/);
1284 }
1285 $self->death("Error: Unterminated pod") unless defined $self->{lastline};
1286 $self->{lastline} = readline($self->{in_fh});
1287 chomp $self->{lastline}; # <-- warning #
1288 $self->{lastline} =~ s/^\s+$//; # <-- warnings 2 #
1289 }
1290 }
#####
1296 sub _maybe_parse_typemap_block {
1297 my ExtUtils::ParseXS $self = shift;
1298
1299 # This is special cased from the usual paragraph-handler logic
1300 # due to the HEREdoc-ish syntax.
1301 if ($self->{lastline} =~ /^TYPEMAP\s*:\s*<<\s*(?:(["'])(.+?)\1|([^\s'"]+?))\s*;?\s*$/) # <-- warning #
1302 {
...
1321 $self->{lastline} = "";
1322 }
1323 }
#####
1418 sub fetch_para {
1419 my ExtUtils::ParseXS $self = shift;
...
1450 if ($self->{lastline} !~ /^\s*#/ # not a CPP directive # <-- warning #
1451 # CPP directives:
1452 # ANSI: if ifdef ifndef elif else endif define undef
1453 # line error pragma
1454 # gcc: warning include_next
1455 # obj-c: import
1456 # others: ident (gcc notes that some cpps have this one)
1457 || $self->{lastline} =~ /^\#[ \t]*
1458 (?:
1459 (?:if|ifn?def|elif|else|endif|elifn?def|
1460 define|undef|pragma|error|
1461 warning|line\s+\d+|ident)
1462 \b
1463 | (?:include(?:_next)?|import)
1464 \s* ["<] .* [>"]
1465 )
1466 /x
...
1469 # Blank line followed by char in column 1. Start of next XSUB?
1470 last if $self->{lastline} =~ /^\S/
1471 && @{ $self->{line} }
1472 && $self->{line}->[-1] eq ""; # <-- warning #
...
1474 # processes CPP conditionals
1475 if ($self->{lastline} # <-- warning #
1476 =~/^#[ \t]*(if|ifn?def|elif|else|endif|elifn?def)\b/)
...
1522 pop(@{ $self->{line} }), pop(@{ $self->{line_no} })
1523 while @{ $self->{line} } && $self->{line}->[-1] eq ""; # <-- warning #
1524
1525 return 1;
1526 }
To assess our test coverage, I created a branch forked from v5.43.3 in which I inserted print statements in dist/ExtUtils-ParseXS/lib/ExtUtils/ParseXS.pm before each of the points where uninitialized value warnings were being emitted when running make for Authen-SASL-XS on FreeBSD-14. I built a threaded perl in the way I customarily do on that platform. I then ran cd t;./perl harness -v ../dist/ExtUtils-ParseXS/t/*.t ; cd -, captured and examined the output. No warnings were emitted at any of the 7 statements indicated by # <-- warning # above.
What I infer from that research is that the .xs files we use as the corpus for ExtUtil-ParseXS's test suite don't exercise the edge cases which are revealed by our investigation into Authen-SASL-XS.
@iabyn (or anyone): How can we extend our test coverage here?