• the setlocal isn't supposed to make a difference (it only applies the command string itself - ie if a file is named 🧆.txt.
• Setting E is more likely to cause this problem then to solve it...

I think these two are related here:

• https://secure.phabricator.com/T13060 (call setlocal in the master php process, so everything is always en_us)
• T15281 and https://secure.phabricator.com/T12071 (Env-vars are missing because of E)

(I'm able to reproduce. Had to apt-get install language-pack-hu for that...)

Is there a Task about this? if not, feel free to open that in Diffusion at least. I'm interested in seeing this fixed.

Ouch sorry renamed again since I forgot the discussion. Still an annoying bug and found your question.

I agree that non-string/null should be handled differently. I guess I don’t see the difference between null + strlen being used vs. the proposed nonempty_string/stringlike, and that making that change is explicitly acknowledging that casting is expected/intentional when it isn’t and instead the different types should be handled appropriately (your suggested long-term solution).

I'm pretty sure that any place where strlen is getting something that isn't a string (including int), that's actually a bug. In Phabricator code, strlen was only supposed to be used when expecting a string or null; All the places we now find aren't strings, are places where we were wrong about the type of the object.
Epriestley says something like that at https://secure.phabricator.com/T13588#257329 .

Honestly I do not love the idea of if(strlen($something)) to answer the question "Is this empty when casted to string?".

Wouldn’t this be better as a null + strlen check? It was originally a strlen I assume.

I don't think I intended to land this change.

Nice! I flagged you as Trusted Contributors. (and if a friend of yours is also able to create such beautiful bug reports, feel free to flag them as well by yourself :D)

Thanks for the quick reply. I updated the question with the backtrace and a shell reproducer

Thanks :) Please try to add more details in order to allow other people to reproduce this problem.

I'm unable to reproduce:
I've deleted a file in arcanist library, and ran arc liberate. I got no error message and the __phutil_library_map__.php was updated correctly.

Yeah it seems the Workflow /project/query/all/customize/?search.objectPHID=PHID-DSHP-<PHID-FOR-REQUEST> has no sense

Premising that now we have a regression here :D

For the records, I had the same experience when deleting a repository via ./bin/remove destroy rESHP for Unknown edge constant "25" (not 26 as mentioned in the task title) in downstream https://phabricator.wikimedia.org/T119588

Oh sorry I love to randomly rename things :D

No, your edit is not correct. The same happens in search, as I indicated in the description:

or craft a query for Diffusion commits, specifying X as author

So if after following my instructions you go to /diffusion/commit/query/all/, you see all of the commits with the author field correctly displayed. If, however, you proceed to editing a query and you select user X as author, you get the "old state".

Let's try to reduce the surface

Done.

Ah sorry for this bug, please visit the files and set Public (I cannot see them)

Just trying to convince you.
If the feature works, it adds functionality that can't be easily replicated. You can have a bunch of complex Herald rules and bypass them all just by declaring that ^temp/ refs are now non-permanent. If the feature doesn't work (or maybe doesn't exist, rather), you have to go to each Herald rule and explicitly exclude ^temp/ branches, which is a lot of work for something that (at least for me) looks like it should be simple.

I'm not, personally, 100% convinced that the hooks should be blocked from running on these refs - there's plenty of edge-cases where this might be even more confusing (commits can become published in a bunch of ways).

Here, I assigned one of your identities to myself:

This is about Diffusion identity assignment feature. List of identities is here: https://we.phorge.it/diffusion/identity/
The simplest way to test it would be to make a test repository and proceed according to the instructions as above. I hope the process is laid out quite clearly.

I'm not quite sure to have understood everything. Can you attach a cute screenshot or something? :)

Patch merged thus resolving

I wonder if this is related to not being able to use the Diffusion repository file auto-complete when not logged in even though the repo is publicly accessible.