1294361 - Firefox translates '%' into "%2525", making displaying of certain files impossible on Gerrit Code Review 2.5.1

Status: RESOLVED INVALID

(Web Compatibility :: Site Reports, defect)

Description

[cornel panceac]

User Agent: Mozilla/5.0 (Windows NT 6.3; WOW64; rv:47.0) Gecko/20100101 Firefox/47.0
Build ID: 20160604131506

Steps to reproduce:

access a Gerrit server to display a file called:

base-files_%.bbappend
	 
Firefox translates the '%' into "%2525" making the file impossible to display. If i change manually the URL to remove 2525, then all is fine.

Comment 1

[cornel panceac]

A similar problem i have for '^' which is translated into "%255E". Example:

project:^yocto.*,n,z

becomes

project:%255Eyocto.*,n,z

, which Gerrit no longer understands.

Comment 2

[Tooru Fujisawa [:arai]]

Can you provide the URL of the page the issue happens and can be reproduced?

Comment 3

[cornel panceac]

https://gerrit.example.com/#/q/status:open+project:^yocto.*,n,z

is translated to

https://gerrit.example.com/#/q/status:open+project:%255Eyocto.*,n,z

Should be reproducible on every Gerrit server

On this first case something interesting thing is happening. The application reports on screen:

Application Error
Server Error
line 1:86 no viable alternative at character '%'
Continue

So maybe this is a problem in the communication between Gerrit and Firefox.
Google Chrome does not have this problem. (But Google created also Gerrit.)

*

https://gerrit.example.com/#/c/29906/2/recipes-core/base-files/base-files_%.bbappend

is translated to 

https://gerrit.example.com/#/c/29906/2/recipes-core/base-files/base-files_%2525.bbappend

So in order to reproduce it you will need to have a Gerrit server 

Gerrit Code Review (2.5.1)

and review a file having '%' in his name, and for the second example, do a search for

project:^yocto.*

I know this is not easy to reproduce unless you are one of the many Gerrit users out there.

Comment 4

[Tooru Fujisawa [:arai]]

Thanks, confirmed the issue.

Here's regression range:
https://hg.mozilla.org/mozilla-central/pushloghtml?fromchange=a38c23ccca0a&tochange=be81b8d6fae9

bug 1093611 might be related, as it touches percent encoding in hash.

Then, the issue doesn't happen on Gerrit Code Review 2.12.3, so I guess they fixed the issue already.
Is it possible to update yours?
Do you have any idea how widely older versions are used?

Comment 5

[cornel panceac]

Honestly i don't know how widely used is this older version, but i would like to never update to the newer version, i completely dislike the new interface (and is not my decision anyway).

Comment 6

[u408661]

Reporter, can you provide a valid public URL that reproduces? It's going to be hard to track down the actual issue without a URL that reproduces. (I tried making a link to a filename with % in it on my local server, and could not reproduce the issue).

Comment 7

[cornel panceac]

No, i do not have a public link for this, but i believe Tooru Fujisava reproduced the issue.

Comment 8

[Jason Duell]

Tooru: see comment 6 and 7.  Do you have a public URI that shows this issue?  We haven't been able to reproduce it so far.

Comment 9

[Tooru Fujisawa [:arai]]

sent URL to nwgh and jduell.

Comment 10

[Tooru Fujisawa [:arai]]

might be better checking how Gerrit fixed the issue on their side.

Comment 11

[Junior [inactive]]

(In reply to Tooru Fujisawa [:arai] from comment #10)
> might be better checking how Gerrit fixed the issue on their side.

Any information for the Gerrit fix?
I've found one bug [1] but its status is still NEW.

[1] https://bugs.chromium.org/p/gerrit/issues/detail?id=3642

Comment 12

[Tooru Fujisawa [:arai]]

Here's bisection result:
  2.12        | good
  2.11        | good
  2.11-rc0    | good
  2.10.7      | bad
  2.10        | bad
  2.9         | bad
  2.5.1       | bad

downloaded from https://gerrit-releases.storage.googleapis.com/index.html

2.11-rc0 sends only one request for the search.
  http://SERVER:PORT/changes/?q=project:^yocto.*&n=25&O=81

2.10.7 sends 2 requests for the search
  http://SERVER:PORT/changes/?q=project:^yocto.*&n=25&O=81
  http://SERVER:PORT/changes/?q=project:%5Eyocto.*&n=25&O=81

there would be some fix or workaround between there.

Comment 13

[Junior [inactive]]

(In reply to Tooru Fujisawa [:arai] from comment #12)
> there would be some fix or workaround between there.

Thanks for doing this.

I've tried but I don't find a related bug fixed in the release note of 2.11.
It just happened to be fixed.

Comment 14

[Tooru Fujisawa [:arai]]

it's unfortunate that I cannot build Gerrit 2.10.7 nor 2.11.0-rc2 locally, while I can build 2.12.4 :/
maybe because of build system incompatibility

it throws error at the start of building:
> Not using buckd because watchman isn't installed.
> Parse error for build file /tmp/gerrit/BUCK:   File "/tmp/gerrit/BUCK", line 3, in <module>
>     gerrit_war(name = 'gerrit')
>   File "/tmp/gerrit/tools/build.defs", line 77, in gerrit_war
>     visibility = visibility,
>   File "/tmp/gerrit/tools/build.defs", line 63, in war
>     visibility = visibility,
>   File "/tmp/gerrit/buck-out/tmp/buck_run.nQmFte/buck6301893443371083208.py", line 130, in invoke
>     return self.func(*args, **updated_kwargs)
> TypeError: genrule() got an unexpected keyword argument 'deps'

I cannot bisect between those versions

Comment 15

[Tooru Fujisawa [:arai]]

I meant 2.11.0-rc0

Comment 16

[Junior [inactive]]

It works if we pretend the user agent is others.
No error prompts.

Comment 17

[cornel panceac]

so you say that the problem is on their side, right? but why google chrome is NOT affected by this problem?  i mean, even if it's their problem, maybe firefox can still work around it ....

Comment 18

[Junior [inactive]]

(In reply to cornel panceac from comment #17)
> so you say that the problem is on their side, right? but why google chrome
> is NOT affected by this problem? 
Since it works when we use other user-agent in the same Firefox, IMO somewhere 
else breaks this use-case.

> i mean, even if it's their problem, maybe firefox can still work around it ....
If there's a more specified behavior (for example what JS behavior makes Gerrit 
escape ^ to %255E), we could find out a way making the world better.

FWIW, it works for later Gerrit per comment 12.

And thanks for reporting this.

Comment 19

[cornel panceac]

Indeed, it was fixed in some newer versions, but i have reported it on the older version, where google chrome does not expose this behaviour. on the other hand, i agree that is not right to put some code in that says if CertainGerritVersion then use theOtherUserAgent :)

Comment 20

[Masatoshi Kimura [:emk]]

Because they sniff the user-agent and do the wrong thing only if the user-agent indicates Firefox.
That said, this bug should have morphed to Tech Evangelism.

Comment 21

[Karl Dubost💡 :karlcow]

(In reply to Junior [:junior] from comment #18)
> FWIW, it works for later Gerrit per comment 12.


Thanks everyone for the analysis. Are the older versions maintained?
If not there will be no hope for outreach and we could close it as WONTFIX or WORKSFORME as it is working in recent versions.

https://bugs.chromium.org/p/gerrit/issues/detail?id=3642#c3

Comment 22

[Mike Taylor [:miketaylr]]

It seems like older versions are not maintained. The solution is to upgrade the older version to the currently working one. Thanks.