I think this is fixed now -- let us know if you're still seeing issues. Thanks for the very thorough report!

I wasn't able to reproduce any hangs or bad behavior from curl locally, but I think D17045 should cause us to select either compression or Content-Length, which is a reasonable thing to expect us to do.

You could also ask Travis for a build.run sort of API. If you're a paying customer of theirs, it might be easier for them to prioritize this request than it is for us to prioritize Travis support. With such an API, there's a fairly straightforward path to Phabricator upstream support.

FWIW, the API seems not intuitive for me: why does the user is able to set arbitrary properties via Conduit, if there is no way to get them back? Is exposing the API actually a hack, to achieve other goals?

Oh, I think the issue is this:

I also can't reproduce this:

Why do you consider this to be a bug, then?

Do you consider this ticket just "make arc patch restore all local commits", or something more?

Oh, that's actually a different task, I don't currently plan to implement that. Let me me rename this to be more clear...

Do we tell you to do this in the documentation anywhere?

This is now better tracked in followups:

This isn't a bug. Diff properties are not intended to be an arbitrary key-value datastore for API consumers.

This field is only populated when you update a revision, and always contains the first line of the update message, which is also posted as a comment.

This hasn't moved forward in a year so I'm just going to close it. We've removed header text in the latest UI iteration, too, and I imagine the button will be replaced completely with universal send-on-enter before long.

Although it's still a little early to be sure, we appear to have fixed the policy issue with unit test failures (D17029). Not sure if that actually impacted this or not.

My dashboard changed at some point and no one else seems to care, so let's wait for this to be more than an @epriestley-problem.

D17037 should fix the documentation, it definitely wasn't very clear as written. That should land some time today and the docs should be updated in about 24 hours.

In the first row of that table, it appears that you have configured a storage.mysql-engine.max-size of 8 ("8 bytes"). Try using 8388608 ("8 megabytes") instead?

If you want to come up with a specific list of domains I'm happy to buy them, but there are many hundreds of domains which differ from phacility.com by only one keystroke and buying hundreds of domains isn't cheap in terms of the amount of work I have to do to do it.

I can find this task by searching for "あああああ" so the CJK stuff may not really be much of an issue anymore? Maybe just an issue with Chinese because the index doesn't know how to tokenize words?

At least today, our error rate on this seems so low that it probably isn't worth implementing. We could take another look at it if we see more of it in the future (perhaps after we begin encouraging application development), but we haven't seen more of this that I can recall recently.

Paste now shows "Old", "New" and "Diff", so you can copy/paste from "Old" to revert. Not perfect, but I think this arises relatively rarely. See T11881 for a followup.

This is extremely old and it's not clear to me which bad cases remain -- I can't immediately find any by fiddling. If any do remain, they haven't turned up again in about two years. If you're still seeing issues, feel free to file a new task with a modern test case.

This picked up a few options through engine support and T11957 is adding some more. Particularly, default view is now "Authored" which I think resolves the original issue. I think the more-root problem is "Dashboards are kind of weird with all the panel copying", which is something we're aware of.

I am very sorry I am declaring t-shirt bankruptcy because I have learned it takes me over five hundred years to ship a t-shirt.

At some point after we get this working for Home, I want to replace the "Quick Create" menu with a similar customizable profile menu, resolving T5867.

rCORE9d89c996844c4599afc1cec9454a71dd80a69c6c threw out the -o pipefail nonsense in CoreHostBackupWorkflow.

(I renamed the "Up for Grabs" column to "Ready", and kicked this out since it doesn't have detailed technical instructions that a new contributor could follow like the other tasks in that column do.)