[tbb-dev] Search Usability Regression in v10

Mon Nov 23 14:42:44 UTC 2020

Matthew, 

Thank you for taking a closer look! It’s interesting that DDG may have changed something on their end – I hadn’t thought of that possibility. 

I did some searching, and it looks like DDG is aware the POST can break back buttons, since they mention it here: https://duckduckgo.com/privacy 
> You can turn on POST requests on our settings page, but it has its own issues. POST requests usually break browser back buttons, and they make it impossible for you to easily share your search by copying and pasting it out of your Web browser's address bar.
> …
> Because of these drawbacks in HTTPS, POST and proxies we decided to take the redirect approach to combat search leakage. However, we leave the choice up to you. You can deviate from the default on our settings page by toggling the redirect or address bar settings. You can also use our encrypted version.

Since the DDG team has documented the limitations of using POST, my intuition is that they’d be unlikely to make changes to help overcome those limitations (e.g., by adjusting the caching settings for search results), but it couldn’t hurt to check with them!

So my preference (at the moment) is also for option (1). 

Thanks!
-Peter 

> On Nov 21, 2020, at 9:42 AM, Matthew Finkel <sysrqb at torproject.org> wrote:
> 
> Hi Peter,
> 
> I looked at this a little closer and this seems like a "regression"
> (change in behavior) on DDG and not in Tor Browser. I tested Tor Browser
> 8.0 (2018) and found the same behavior.
> 
> I suspect this usability regression is caused by the results page not
> being considered cachable. Normally, Tor Browser (Firefox) would save
> the previous page in the bfcache (back-forward cache) and re-use it when
> the Back button is pressed, but this is not happening now and the page
> is requested again instead.
> 
> There are three obvious paths forward.
> 
> 1) Use a GET request instead of POST
>  - This does not resolve the regression, but when the page is
>    re-requested on "Back" the original search terms are included in the
>    GET request and the user is presented with the results page again
> 
> 2) Results from html.duckduckgo.com are (apparently) successfully cached
>   (currently used when the Safest security level is set in Tor Browser)
>  - Use this domain on all security levels, instead of the main DDG site
>    with Javascript optimizations
> 
> 3) Work with DDG and identify why this is broken and find a resolution
> 
> 
> I am leaning toward (1) because:
>  a) This is the default method in Firefox
>  b) I am skeptical about the (security) benefits of POST over GET
>  c) We know this works
> 
> We'll get some complaints about switching to using GET as the default
> search method in Tor Browser, so I'll reach out to DDG and work with
> them on this, too.
> 
> I'd like to hear concerns about switching to using GET, too.
> 
> On Wed, Nov 18, 2020 at 04:36:33PM -0500, Peter Story wrote:
>> Hello Matthew, 
>> 
>> Thanks for the quick reply! And no worries at all. 
>> 
>> Glad to hear that the team is aware of the problem! I think that switching back to GETs (I believe this was the behavior before v10) wouldn’t significantly increase the threat of should surfing. With the current configuration, the search is still visible on the search results page itself, and after clicking on a linked page the text of the search is still visible in the browser’s back button. 
>> 
>> I think this regression has the potential to seriously impact the browser’s usability, because when following links in search, a significant percent of linked websites block Tor users. When a user encounters a page which blocks them, they will probably try to return to the search results to try another website. Crucially, the back button not returning to the search results breaks this workflow.
>> 
>> How does decision-making for issues like this typically work? If I’d like to help fix this issue, do you have any recommendations (e.g., should I open a PR with a fix)? 
>> 
>> Thank you!
>> -Peter 
>> 
>>> On Nov 18, 2020, at 4:20 PM, Matthew Finkel <sysrqb at torproject.org> wrote:
>>> 
>>> On Wed, Nov 18, 2020 at 03:59:15PM -0500, Peter Story wrote:
>>>> I reported this usability problem on the tor-talk list but received no replies, so I thought I’d try here. For some context, I’m a researcher at Carnegie Mellon University, and I’m considering running a user-study involving Tor Browser in the coming months. I think this usability problem might negatively impact users (in my study, and at large), so I thought I should bring it to your attention. 
>>>> 
>>> 
>>> Hi Peter,
>>> 
>>> Thanks for sending this mail and following up. I apologize for the lack
>>> of response, we're all overloaded.
>>> 
>>> This is a known usability issue. It arises from the fact that Tor
>>> Browser sends the initial search query via a POST request. The breakage
>>> occurs because the "Back" action triggers a GET request for the previous
>>> site and it does not re-play the original POST request. I assume the
>>> reasoning for this is that POST requests are not assumed to be
>>> idempotent, whereas GET requests are. I'm not sure how best we can solve
>>> this except switching to using GET, however this places all queries into
>>> the URL bar and then they become vulnerabilbe to shoulder-surfing as
>>> such. That probably a necessary trade-off, though.
>>> 
>>>> 
>>>>> I’m seeing a small but annoying usability problem in v10 of Tor Browser (tested with 10.0.2 on a Mac). Replication:
>>>>> 
>>>>> Open a new window
>>>>> Search using either the URL bar or the “Search with DDG” field. On the page with search results, note that the page’s URL doesn’t contain the search query in the URL parameters.
>>>>> Click on a link in the search results
>>>>> Click the back button
>>>>> Actual behavior: you are returned to a blank search page
>>>>> Expected behavior: you are returned to your search results
>>>>> 
>>>>> I don’t see this problem in the latest version of Firefox, 82.0.3, and I don’t remember seeing this problem before Tor Browser v10. Also, I do not see this problem for searches conducted with Google in Tor Browser. 
>>>>> 
>>>>> I’m wondering whether this was an intentional change, or something that should be fixed. Hopefully it's an easy fix: if the search is conducted using URL parameters, then the back button should work as expected. 
>>>> 
>>>> Sincerely,
>>>> -Peter Story
>>>> PhD Student
>>>> 
>>> 
>>>> _______________________________________________
>>>> tbb-dev mailing list
>>>> tbb-dev at lists.torproject.org <mailto:tbb-dev at lists.torproject.org>
>>>> https://lists.torproject.org/cgi-bin/mailman/listinfo/tbb-dev <https://lists.torproject.org/cgi-bin/mailman/listinfo/tbb-dev>
>>> 
>>> _______________________________________________
>>> tbb-dev mailing list
>>> tbb-dev at lists.torproject.org <mailto:tbb-dev at lists.torproject.org>
>>> https://lists.torproject.org/cgi-bin/mailman/listinfo/tbb-dev <https://lists.torproject.org/cgi-bin/mailman/listinfo/tbb-dev>
> 
>> _______________________________________________
>> tbb-dev mailing list
>> tbb-dev at lists.torproject.org
>> https://lists.torproject.org/cgi-bin/mailman/listinfo/tbb-dev
> 
> _______________________________________________
> tbb-dev mailing list
> tbb-dev at lists.torproject.org
> https://lists.torproject.org/cgi-bin/mailman/listinfo/tbb-dev

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.torproject.org/pipermail/tbb-dev/attachments/20201123/7dd561b5/attachment.htm>