607217 - Submitting a comment on tor.com is impossible

Status: RESOLVED FIXED

(Tech Evangelism Graveyard :: English US, defect)

Description

[Josh Matthews [:jdm]]

STR:
1. Go to a post on tor.com
2. Log in to an existing account
3. Leave a comment
4. Confirm comment posting

Expected:
Comment is posted.  These steps work fine in the latest dev Chrome build.

Actual:
"500 - View not found [name, type, prefix]: frontpage,html,commentView" error page displayed.

Comment 1

[Josh Matthews [:jdm]]

I'll attempt to get an HTTP log for this behaviour, and find a regression range.  I've checked the latest nightly and beta 6, and they both display the error page.

Comment 2

[Josh Matthews [:jdm]]

Attachment: Failing http/sockettransport/hostresolve log

Comment 3

[Josh Matthews [:jdm]]

3.6.9 saves the comment correctly.

Comment 4

[Josh Matthews [:jdm]]

Attachment: Successful http/sockettransport/hostresolve log

Comment 5

[Josh Matthews [:jdm]]

Comparing the logs, the nightly seems to go straight from the final POST to receiving a 500 error, while 3.6.9 receives a 303 See Other.  I haven't found any smoking bullets, unfortunately.

Comment 6

[Josh Matthews [:jdm]]

Moving this to Core for the foreseeable future.

Comment 7

[Josh Matthews [:jdm]]

4.0b1 is also broken.

Comment 8

[Josh Matthews [:jdm]]

3.6.11 works.

Comment 9

[Josh Matthews [:jdm]]

As does 3.7a2.

Comment 10

[Josh Matthews [:jdm]]

By which I mean that I tested the wrong build.  The most definitive regression range to date is that it worked in 3.7a1 and broke in 3.7a2.

Comment 11

[Josh Matthews [:jdm]]

Bisecting is narrowing down the cause to sicking's pushes in bugs bug 544698 and bug 546528.

Comment 12

[Josh Matthews [:jdm]]

I was overzealous; apparently this is the fault of bug 547165 instead. I'm working on a reduced testcase now.

Comment 13

[Josh Matthews [:jdm]]

So it turns out that this is due to the change where <input type="image" name="foo" value="bar"> no longer submits foo=bar.  Tor.com is using that value (in this case, task=post/cancel/preview) server-side to determine what action to take, and throws a 500 on no value submitted (actually they get task="" because there's also a hidden element named task for some reason).

I threw together a testcase which is composed of the tor.com comment form which submits to a PHP script that prints out $_POST.  FWIW, Chrome submits task=post.

Comment 14

[Josh Matthews [:jdm]]

Testcase at http://www.joshmatthews.net/bug607217/ in case it's required.

Comment 15

[Boris Zbarsky [:bzbarsky]]

IE doesn't submit that, does it?  And the HTML spec requires not submitting it...

Does IE work on the tor page in question?

Comment 16

[Johnny Stenback (:jst)]

Mounir, could you have a look here?

Comment 17

[Mounir Lamouri (:mounir)]

IE and Opera have task="" and both work perfectly. I tried to make Opera identify itself like Firefox and it was still working so I'm not sure if there is UA sniffing or something else.

Comment 18

[Josh Matthews [:jdm]]

That's interesting.  Do they still actually submit task="", or is it not present at all like FF?

Comment 19

[Mounir Lamouri (:mounir)]

(In reply to comment #18)
> That's interesting.  Do they still actually submit task="", or is it not
> present at all like FF?

Task has no value, like Firefox.

Comment 20

[Boris Zbarsky [:bzbarsky]]

Er...  Opera submits a "task=" on Josh's minimal testcase, no?  So presumably submits task=whatever on the tor site?

What does IE do on Josh's minimal testcase?  Does it end up with |[task] =>| in the output?

Comment 21

[Boris Zbarsky [:bzbarsky]]

Oh, nevermind.  Opera and we both send |[task] =>|...

What do the various browsers send on the actual tor site?  Do we have any contacts with the tor folks?

Comment 22

[Boris Zbarsky [:bzbarsky]]

I'd just mail the tor folks, but it's hard to do that without having some idea of what their actual stuff looks like when they reply.  jdm, are you willing to do that and ask them about the behavior difference between Fx4 and Opera here?

Comment 23

[Josh Matthews [:jdm]]

Since the task value comes from the button that's used to submit, whatever you
see on the testcase should be identical to the tor.com form.

And yeah, I'll try and get in contact with them.

Comment 24

[Mounir Lamouri (:mounir)]

Ok, I've found what's happening and it explains why we have this <input type='hidden' name="task" value=""/>

In the preview page, for IE and Opera, we have this:

<script type="text/javascript">

  document.write('<input type="image" src="/templates/tor/images/buttons/button-preview-comment.gif" value="preview" class="tor_subm" onclick="this.form.task.value=\'preview\';" />' +

  '<input type="image" src="/templates/tor/images/buttons/button-post.gif" value="post" class="tor_subm" onclick="this.form.task.value=\'post\';" />' +

  '<input type="image" src="/templates/tor/images/buttons/button-cancel.gif" value="cancel" class="tor_subm" onclick="this.form.task.value=\'cancel\';" />');

</script>

<noscript>

  <input type="submit" name="task" value="preview" class="tor_subm" />

  <input type="submit" name="task" value="post" class="tor_subm" />			

  <input type="submit" name="task" value="cancel" class="tor_subm" />			

</noscript>

instead of this (Webkit and Gecko):

<input type="image" src="/templates/tor/images/buttons/button-preview-comment.gif" name="task" value="preview" class="tor_subm" />
<input type="image" src="/templates/tor/images/buttons/button-post.gif" name="task" value="post" class="tor_subm" />
<input type="image" src="/templates/tor/images/buttons/button-cancel.gif" name="task" value="cancel" class="tor_subm" />

Comment 25

[Mounir Lamouri (:mounir)]

So maybe this bug should be assigned to Evangelism and shouldn't be a blocker anymore?

Comment 26

[Boris Zbarsky [:bzbarsky]]

Yeah, indeed.  This just looks like browser-sniffing brokenness, since the "IE and Opera" codepath would work just fine for Webkit/Gecko.

Someone still needs to let tor know that their code is broken.

Comment 27

[Mounir Lamouri (:mounir)]

(I guess "regression" and "testcase" do not really make sense here)

Josh, I can send an email now if you didn't do it.

Comment 28

[Josh Matthews [:jdm]]

I have not started the process, so feel free.

Comment 29

[Mounir Lamouri (:mounir)]

Just sent an email to [e-mail]

Comment 30

[Mounir Lamouri (:mounir)]

They did react very quickly. This seems fixed now.

Comment 31

[Jonas Sicking (:sicking) No longer reading bugmail consistently]

Yay! Thanks Mounir