feat(chat): support user shell env in shell command tool

[GoodluckH]

Issue #, if available:
#1030

Description of changes:

• Renamed execute_bash tool to execute_shell_commands
• Fetches all aliases based on Shell type (currently support zsh and bash) from user config files

By submitting this pull request, I confirm that you can use, modify, copy, and redistribute this contribution, under the terms of your choice.

[codecov-commenter]

Codecov Report

❌ Patch coverage is 1.68067% with 117 lines in your changes missing coverage. Please review.
✅ Project coverage is 13.69%. Comparing base (e0f65f3) to head (ecce6f7).
⚠️ Report is 610 commits behind head on main.

Files with missing lines	Patch %	Lines
...q_cli/src/cli/chat/tools/execute_shell_commands.rs	0.91%	108 Missing ⚠️
crates/q_cli/src/cli/chat/tools/mod.rs	0.00%	9 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #1043      +/-   ##
==========================================
- Coverage   13.69%   13.69%   -0.01%     
==========================================
  Files        2361     2361              
  Lines      203706   203745      +39     
  Branches   184070   184109      +39     
==========================================
+ Hits        27894    27895       +1     
- Misses     174425   174462      +37     
- Partials     1387     1388       +1     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.

[mschrage]

We should consider launching the process using --login and --interactive to emulate the behavior of the user's shell.

Pros:

• Will source relevant dotfiles automatically.

Cons:

• May add additional latency to shell execution as all dotfiles need to be sourced
• Introduces environment dependency which can result in inconsistent behavior that is difficult to debug.

[GoodluckH]

Yes, I initially tried using interactive mode to source config files but it will result in tty suspension error that will kill the Q CLI session.

I'm still investigating how to avoid that.

[mschrage]

Some interesting discussion here:

This will fix it:

stty -tostop
From the man page:

tostop (-tostop)
Send (do not send) SIGTTOU for background output. This causes background jobs to stop if they attempt terminal output.
This tostop is normally the default setting, as it's usually undesirable to mix the output of

https://stackoverflow.com/questions/24056102/why-do-i-get-suspended-tty-output-in-one-terminal-but-not-in-others

[GoodluckH]

Yes I tried all those but didn't work.

Instead, I will launch a PTY.

[jsamuel1]

zsh, bash, fish.
What other ones are supported in other parts of q cli?

How will automated tests work for this? Docker based, or can you do it with chroot/mocks?

[brandonskiser]

This should be a workspace dependency

[brandonskiser]

Can you make this a workspace dependency and update figterm as well? (figterm already depends on 0.8)

[brandonskiser]

nit - move this after all of the imports, it's declaring a new function, also update the name to something more readable preferably, like set_terminal_winsize or something

[brandonskiser]

Since the above lines are copied straight from figterm (and they're particularly complicated) I wonder if this could be generalized in a library, e.g. in fig_util

[brandonskiser]

This doesn't work with fish, we'd need to get the current shell and fallback to bash otherwise

also, probably best to use https://github.com/aws/amazon-q-developer-cli/blob/main/crates/fig_util/src/shell.rs#L90

[brandonskiser]

This method seems to break our (admittedly already poor support of) interactive commands, e.g. npx create-react-app helloworld

My thought was that we would spawn the child with a pty, and set our terminal to raw mode. This way, we can copy all bytes directly to the child which can then handle sigint etc. which should support more interactive commands.

[brandonskiser]

removing stderr?

[GoodluckH]

With the PTY mode it's hard to differentiate stderr, but we can already infer from status code if the resulting message is error or not. The output of this function is to give some context to the LLM to inform it of the execution result.

[brandonskiser]

I was also thinking about how we can handle interactive commands since some (e.g. vim) shouldn't have their stdout returned back to the model, still an open question to solve