CLT UI cosmetic touches and issues

[donhardman]

There are several issues that are not very critical but are great to address when we can:

• When we land with branch=.. we should reset the query string, because the branch has already switched, and reloading the page will reswitch it again.
• The "Create PR" button unnecessarily triggers redrawing and animation on every git status fetch; that should only be done once.
• It is not possible to scroll the Ask AI output; when we scroll, we should stop snapping to the end.
• If we have a rep file, it would be good to render the LAST test results when we open it instead of pending results.
• When we switch to a branch, the PR is not switched to the commit; it only happens once we click there, which creates some confusion.
• There is an issue with loading "named" sessions in ASK AI. We should consider removing named sessions for simplicity or to fix the issue.
• Instead of using timestamps in session names, we should use Y-m-d-H-i format for more clarity.
• New session should create session without asking for a custom name

[sanikolaev]

• When I create a new session let's hide this completely (or make it much smaller) until it starts processing the request:
  
  Otherwise it's inconvenient to provide a long prompt, e.g.:

  Write a test which does this:
  
  1. starts Manticore
  
  2. Loads some data into it running selects too while loading:
  manticore-load \
  --host=127.0.0.1 \
  --drop --batch-size=1000 --threads=4 --total=1000000 \
  --init="CREATE TABLE test(id bigint, name text, type int)" \
  --load="INSERT INTO test(id,name,type) VALUES(<increment>,'<text/10/100>',<int/1/100>)" \
  --together \
  --threads=1 --total=5000 \
  --load="SELECT * FROM test WHERE MATCH('<text/1/1>')"
  
  3. then runs mysql -P9306 -h0 -e "show table test status like '%search_stats_%'; show status like 'search_stats%'"
  
  and ensures that the global stats do not differ much from the per-table stats (not more than by 20%)
  

• When I pull this down the window closes:
  

• If I press "Create PR" and then "Cancel" it becomes "Commit":
  
  
  After refreshing the page it becomes "Create PR" back.

• I suggest to simplify this window:
  

• The button disappears when I mouse over it:
  
  Let's also change the button name to ⬆️ or smth as it seems to be a standard for submitting a prompt (chatgpt, deepseek, z.ai) and "Send" is confusing anyway.

[sanikolaev]

• I can't continue a timed out session:
  
  If I submit "Continue", it doesn't remember the previous context.
• There's a discrepancy in the amounts for the same session:
  

[sanikolaev]

• docker pull is not run automatically before each test run. It's inconvenient to run it manually on the server (which helps BTW)
• after AI updates a test, it's not refreshed until I refresh the whole page
• AI can create regexes for a single test step, but writing a prompt for it is a bit of a hassle. It would be much easier if there was a special button for that — I could just click it, wait a minute, and get the regex ready.
• I suggest we improve the PR template so:
  

[sanikolaev]

• cmd-z doesn't work when entering a command which is inconvenient
• In this case a PR already existed, but it didn't detect it and suggested to create a new one instead of just commiting to the existing PR:
  
  As discussed, it has smth to do with the user's token being expired
• I can't see a raw test, so if something happens I can't manually copy the test contents to my editor to commit it manually
• https://clt.manticoresearch.com/?test_path=http-interface%2Fcli-endpoint.rec opens
  
  Expected: the test opens right away.
• Let's grey out the whole test when it's running, so it's easier to notice when it's finished. Noticing when this little text disappears is difficult:
  
• Sometimes it would be convenient to discard just one command:
  

[sanikolaev]

• we need a way to specify a default docker image (or a list of them) in a comment in a test. Advantages:
  • no need to change it manually in the UI
  • we can simplify using CLT in GitHub CI. If we support a list of images we can even try to get rid of the matrices.

[donhardman]

• We should NOT allow running tests with empty commands or other blocks because it leads to unexpected results; the UI should report an error.

[donhardman]

• Unable to switch to branch when we have a FILE with the SAME name (new one) and the branch we're switching to ALSO has changes to this file, so this goes to conflict. Instead we should RESET and CLEAN -fdx

[PavelShilin89]

Tests containing a block to launch dind at startup only Failed for all commands, of course.

[donhardman]

• As discussed, it's good to move to sparse checkout on a per-branch basis per user, so in that case each branch
  will be used as a separate folder to run tests. In that case, we can use multiple tabs to run tests.
• Also, as improvements to the UI, we should add branch autocomplete and move it to the top left, doing it almost
  the same way GitHub's UI does.
• We should add the possibility to add a description and implement a way to set "Default image:" for running
  tests on a given Docker image. For defaults in input, we should remove them from the header and use the default if
  specified, only using them when the user specifically edits the input.

[donhardman]

• Investigate and fix the issue with resuming sessions. It seems there is a bug in the external tool we use for this.

[sanikolaev]

• this is confusing. I don't know what it means:

[sanikolaev]

• It looks wrong and confusing that intermediate test steps are not complete, but later steps are done:

[donhardman]

I'm closing the issue due to most things being fixed and some things that are big enough extracted into new issues:

• Feature: Branch per workdir on user #106
• Feature: advanced configuration #107
• Feature: auto refine #108