Jeff King [off-list ref] writes:
On Sat, Jun 09, 2012 at 08:53:48PM +0200, Javier.Roucher-Iglesias@ensimag.imag.fr wrote:
quoted
diff --git a/contrib/mw-to-git/git-remote-mediawiki b/contrib/mw-to-git/git-remote-mediawiki
index c18bfa1..4b14d78 100755
--- a/contrib/mw-to-git/git-remote-mediawiki
+++ b/contrib/mw-to-git/git-remote-mediawiki
@@ -152,28 +152,111 @@ while (<STDIN>) {
########################## Functions ##############################
# MediaWiki API instance, created lazily.
+sub run_credential {
Is there any reason not to add this to perl/Git.pm? I suspect that other
scripts will want to use it, too (for example, send-email could probably
use it for SMTP credentials).
Currently, git-remote-mediawiki is a standalone script (doesn't use
Git.pm). This is good because it makes it trivial to install, but bad in
the sense that it may force us (or others) to reinvent the wheel.
Until now, the wheels we reinvented were very simple (run_git
essentially), but we may be reaching the point where it makes sense to
use and contribute to Git.pm.
Unfortunately, from a non-technical point of view, Javier is
contributing this as part of a student project, which ends this week,
and it's probably not reasonable to introduce such change so late. So,
I'd keep it here at least for now, and a move to Git.pm could be a
separate future topic.
quoted
+ if (scalar(@_) == 2) {
+ if ($_[1] eq ("store" || "cache")) {
+ run_git("config credential.helper \'$_[1]\'");
+ } else {
+ print STDERR "ERROR: run_credential (fill|approve|reject) [store|cache]\n";
+ exit 1;
+ }
+ }
This hunk looks wrong. You should never be setting the credential.helper
config; that is the responsibility of the user to set, as they want to
select whatever helper is appropriate. Nor do you need to care about
which helpers are in use; the point of git-credential is that it will do
that for you.
Absolutely.
sub fill_credential {
my $quoted_url = quotemeta(shift);
my $verbatim = `git credential fill $quoted_url`;
$? and die "git-credential failed";
$verbatim =~ /^username=(.*)$/m
or die "git-credential did not give us a username";
my $username = $1;
$verbatim =~ /^password=(.*)$/m
or die "git-credential did not give us a password";
return ($username, $password, $verbatim);
}
sub report_credential {
my ($type, $verbatim) = @_;
open(my $fh, '|-', "git credential $type");
print $fh $verbatim;
}
That sounds sensible too. We should be careful not to give a password as
argument (or users of the same machine will be able to find it with e.g.
"ps u"), but your proposal is OK with that.
quoted
+ # error if key undef
+ if (not defined $key) {
+ print STDERR "ERROR reciving reponse git credential fill\n";
+ exit 1;
+ }
[...]
quoted
+ } else {
+ while (<Reader>) {
+ print STDERR "\nERROR while running git credential $op:\n$_";
+ }
+ }
+}
This isn't a good way to check for errors. The non-fill actions will
never produce output on stdout, and you are not intercepting their
stderr. Besides which, checking for errors by reading stderr is not a
good practice; you should check the return value of the command in $?
after it finishes.
I think it should do both. In case "git credential fill" returns
something that doesn't match the regexp, we don't want perl to error
with "use of undefined value", but that's just being defensive because
it shouldn't happen.
--
Matthieu Moy
http://www-verimag.imag.fr/~moy/
On Sun, 10 Jun 2012 15:37:42 +0200, Matthieu Moy wrote:
Jeff King [off-list ref] writes:
quoted
On Sat, Jun 09, 2012 at 08:53:48PM +0200,
Javier.Roucher-Iglesias@ensimag.imag.fr wrote:
quoted
diff --git a/contrib/mw-to-git/git-remote-mediawiki
b/contrib/mw-to-git/git-remote-mediawiki
index c18bfa1..4b14d78 100755
--- a/contrib/mw-to-git/git-remote-mediawiki
+++ b/contrib/mw-to-git/git-remote-mediawiki
@@ -152,28 +152,111 @@ while (<STDIN>) {
########################## Functions ##############################
# MediaWiki API instance, created lazily.
+sub run_credential {
Is there any reason not to add this to perl/Git.pm? I suspect that
other
scripts will want to use it, too (for example, send-email could
probably
use it for SMTP credentials).
Currently, git-remote-mediawiki is a standalone script (doesn't use
Git.pm). This is good because it makes it trivial to install, but bad
in
the sense that it may force us (or others) to reinvent the wheel.
Until now, the wheels we reinvented were very simple (run_git
essentially), but we may be reaching the point where it makes sense
to
use and contribute to Git.pm.
Unfortunately, from a non-technical point of view, Javier is
contributing this as part of a student project, which ends this week,
and it's probably not reasonable to introduce such change so late.
So,
I'd keep it here at least for now, and a move to Git.pm could be a
separate future topic.
Thank's for explain my situation.
quoted
quoted
+ if (scalar(@_) == 2) {
+ if ($_[1] eq ("store" || "cache")) {
+ run_git("config credential.helper \'$_[1]\'");
+ } else {
+ print STDERR "ERROR: run_credential (fill|approve|reject)
[store|cache]\n";
+ exit 1;
+ }
+ }
This hunk looks wrong. You should never be setting the
credential.helper
config; that is the responsibility of the user to set, as they want
to
select whatever helper is appropriate. Nor do you need to care about
which helpers are in use; the point of git-credential is that it
will do
that for you.
Absolutely.
I have add this with no advance warning, but i will remove it in the
next patch.
quoted
sub fill_credential {
my $quoted_url = quotemeta(shift);
my $verbatim = `git credential fill $quoted_url`;
$? and die "git-credential failed";
$verbatim =~ /^username=(.*)$/m
or die "git-credential did not give us a
username";
my $username = $1;
$verbatim =~ /^password=(.*)$/m
or die "git-credential did not give us a
password";
return ($username, $password, $verbatim);
}
sub report_credential {
my ($type, $verbatim) = @_;
open(my $fh, '|-', "git credential $type");
print $fh $verbatim;
}
That sounds sensible too. We should be careful not to give a password
as
argument (or users of the same machine will be able to find it with
e.g.
"ps u"), but your proposal is OK with that.
quoted
quoted
+ # error if key undef
+ if (not defined $key) {
+ print STDERR "ERROR reciving reponse git credential fill\n";
+ exit 1;
+ }
[...]
to be change, thanks for the corrections
quoted
quoted
+ } else {
+ while (<Reader>) {
+ print STDERR "\nERROR while running git credential $op:\n$_";
+ }
+ }
+}
This isn't a good way to check for errors. The non-fill actions will
never produce output on stdout, and you are not intercepting their
stderr. Besides which, checking for errors by reading stderr is not
a
good practice; you should check the return value of the command in
$?
after it finishes.
I think it should do both. In case "git credential fill" returns
something that doesn't match the regexp, we don't want perl to error
with "use of undefined value", but that's just being defensive
because
it shouldn't happen.
On Sun, Jun 10, 2012 at 03:37:42PM +0200, Matthieu Moy wrote:
quoted
Is there any reason not to add this to perl/Git.pm? I suspect that other
scripts will want to use it, too (for example, send-email could probably
use it for SMTP credentials).
Currently, git-remote-mediawiki is a standalone script (doesn't use
Git.pm). This is good because it makes it trivial to install, but bad in
the sense that it may force us (or others) to reinvent the wheel.
Until now, the wheels we reinvented were very simple (run_git
essentially), but we may be reaching the point where it makes sense to
use and contribute to Git.pm.
Yeah, I noticed that. But hopefully since they are at least in the same
distribution, it is just a matter of getting the Makefiles right, and
will not be an extra burden on the user.
Unfortunately, from a non-technical point of view, Javier is
contributing this as part of a student project, which ends this week,
and it's probably not reasonable to introduce such change so late. So,
I'd keep it here at least for now, and a move to Git.pm could be a
separate future topic.
Totally understood. I review from git's perspective, but it is up to you
to manage your students' workload. We can migrate the code to Git.pm
later (probably as part of a series which actually introduces a second
caller).
quoted
sub fill_credential {
my $quoted_url = quotemeta(shift);
my $verbatim = `git credential fill $quoted_url`;
$? and die "git-credential failed";
$verbatim =~ /^username=(.*)$/m
or die "git-credential did not give us a username";
my $username = $1;
$verbatim =~ /^password=(.*)$/m
or die "git-credential did not give us a password";
return ($username, $password, $verbatim);
}
sub report_credential {
my ($type, $verbatim) = @_;
open(my $fh, '|-', "git credential $type");
print $fh $verbatim;
}
That sounds sensible too. We should be careful not to give a password as
argument (or users of the same machine will be able to find it with e.g.
"ps u"), but your proposal is OK with that.
Yes, that was intentional (and is the reason why helpers do so much over
stdin, even though it would reduce their parsing load if they could use
command-line arguments). Unfortunately, there is still one case that
reveals a password on the command-line: if a caller has a URL that
contains an embedded password like:
https://bob:secret@example.com/foo.git
It's tempting to say "well, then they don't need to ask git-credential
at all!". But the point of handing the whole URL to git-credential is so
that the caller doesn't _have_ to do the parsing. So how would it know
that the URL contains a password? :)
So instead of a URL on the command-line, it might make sense to simply
let the caller send an extra "url=" parameter on stdin (in addition to
any broken-down parameters, if it also wishes). It's way less convenient
(you are stuck with open2 from perl, rather than simple backticks), but
I think we should be cautious due to the security implications.
quoted
quoted
+ # error if key undef
+ if (not defined $key) {
+ print STDERR "ERROR reciving reponse git credential fill\n";
+ exit 1;
+ }
[...]
quoted
quoted
+ } else {
+ while (<Reader>) {
+ print STDERR "\nERROR while running git credential $op:\n$_";
+ }
+ }
+}
This isn't a good way to check for errors. The non-fill actions will
never produce output on stdout, and you are not intercepting their
stderr. Besides which, checking for errors by reading stderr is not a
good practice; you should check the return value of the command in $?
after it finishes.
I think it should do both. In case "git credential fill" returns
something that doesn't match the regexp, we don't want perl to error
with "use of undefined value", but that's just being defensive because
it shouldn't happen.
Sorry, I just meant the latter block. It is checking for non-fill
actions to send any output, which they will never do (yes, it would be
an error for them to do so, but it is a very unlikely bug, and IMHO not
really worth checking).
-Peff