CC'ing ports@ in case people miss this on bugs@. Quick summary from
earlier posts: some mirrors e.g. leaseweb, esc7, have directory listings
with <a href="foo" title="foo">, which pkg_add's parser can't handle.
On 2017/08/03 20:44, Taylor Stearns wrote:
> On Wed, Aug 02, 2017 at 11:41:52PM +0100, Stuart Henderson wrote:
> > Please send a diff -u and don't indent it.
>
> Here is the CVS diff -u:
Thanks, that's readable now :)
Normally I'd want to wait until espie@ is around for pkg_add changes
but the leaseweb mirror is at the top of ftplist.cgi when fetched over
https from a "new" IP address (it doesn't currently prioritize nearby
mirrors like it does over http) so it's somewhat likely that fresh
installs will be using this mirror by default, so it would be nice
to get a fix committed sooner..
Generally I agree with the diff except for one thing (below)
> Index: PackageRepository.pm
> ===================================================================
> RCS file: /cvs/src/usr.sbin/pkg_add/OpenBSD/PackageRepository.pm,v
> retrieving revision 1.145
> diff -u -p -r1.145 PackageRepository.pm
> --- PackageRepository.pm 20 Jun 2017 18:05:44 -0000 1.145
> +++ PackageRepository.pm 3 Aug 2017 19:27:55 -0000
> @@ -912,7 +912,7 @@ sub get_http_list
> $error) or return;
> while(<$fh>) {
> chomp;
> - for my $pkg (m/\<A\s+HREF=\"(.*?\.tgz)\"\>/gio) {
> + for my $pkg (m/\<A\s+HREF=\"(.*?\.tgz)\"/gio) {
> $pkg = $1 if $pkg =~ m|^.*/(.*)$|;
> # decode uri-encoding; from URI::Escape
> $pkg =~ s/%([0-9A-Fa-f]{2})/chr(hex($1))/eg;
> Index: PackageRepository/HTTP.pm
> ===================================================================
> RCS file: /cvs/src/usr.sbin/pkg_add/OpenBSD/PackageRepository/HTTP.pm,v
> retrieving revision 1.12
> diff -u -p -r1.12 HTTP.pm
> --- PackageRepository/HTTP.pm 27 Aug 2014 18:40:03 -0000 1.12
> +++ PackageRepository/HTTP.pm 3 Aug 2017 19:27:55 -0000
> @@ -280,7 +280,7 @@ sub get_directory
> exit 1;
> }
> print "SUCCESS: directory $dname\n";
> - for my $pkg ($r =~ m/\<A\s+HREF=\"(.+?)\.tgz\"\>/gio) {
> + for my $pkg ($r =~ m/\<A\s+HREF=\"(.+?)\.tgz\"/gio) {
> $pkg = $1 if $pkg =~ m|^.*/(.*)$|;
> # decode uri-encoding; from URI::Escape
> $pkg =~ s/%([0-9A-Fa-f]{2})/chr(hex($1))/eg;
>
The href doesn't have to immediately follow the "<a ", and it's easy
enough to cope, and I can't think of a downside, so there's a modified
diff below that handles this.
Any OKs for either Taylor's or my amended version?
I've tested this with "pkg_add moo" from snapshot packages from *all*
http mirrors, it works on all the non-broken ones (failures at unbr.br
which doesn't use --delete, and ulisboa.pt which is down from my
network, but these are obviously not a problem with the diff).
Index: OpenBSD/PackageRepository.pm
===================================================================
RCS file: /cvs/src/usr.sbin/pkg_add/OpenBSD/PackageRepository.pm,v
retrieving revision 1.145
diff -u -p -r1.145 PackageRepository.pm
--- OpenBSD/PackageRepository.pm 20 Jun 2017 18:05:44 -0000 1.145
+++ OpenBSD/PackageRepository.pm 3 Aug 2017 21:04:28 -0000
@@ -912,7 +912,7 @@ sub get_http_list
$error) or return;
while(<$fh>) {
chomp;
- for my $pkg (m/\<A\s+HREF=\"(.*?\.tgz)\"\>/gio) {
+ for my $pkg (m/\<A[^>]*\s+HREF=\"(.*?\.tgz)\"/gio) {
$pkg = $1 if $pkg =~ m|^.*/(.*)$|;
# decode uri-encoding; from URI::Escape
$pkg =~ s/%([0-9A-Fa-f]{2})/chr(hex($1))/eg;
Index: OpenBSD/PackageRepository/HTTP.pm
===================================================================
RCS file: /cvs/src/usr.sbin/pkg_add/OpenBSD/PackageRepository/HTTP.pm,v
retrieving revision 1.12
diff -u -p -r1.12 HTTP.pm
--- OpenBSD/PackageRepository/HTTP.pm 27 Aug 2014 18:40:03 -0000 1.12
+++ OpenBSD/PackageRepository/HTTP.pm 3 Aug 2017 21:04:28 -0000
@@ -280,7 +280,7 @@ sub get_directory
exit 1;
}
print "SUCCESS: directory $dname\n";
- for my $pkg ($r =~ m/\<A\s+HREF=\"(.+?)\.tgz\"\>/gio) {
+ for my $pkg ($r =~ m/\<A[^>]*\s+HREF=\"(.+?)\.tgz\"/gio) {
$pkg = $1 if $pkg =~ m|^.*/(.*)$|;
# decode uri-encoding; from URI::Escape
$pkg =~ s/%([0-9A-Fa-f]{2})/chr(hex($1))/eg;
