Make gpt-engineer pip installable/runnable

[Jeadie]

Feature

Make it easy to run

pip install .  # eventually pip install gpt-engineer
gpt-engineer <PATH> # or gpt-engineer --help

Changeset

• improve pyproject.toml with project configs
• make gpt_engineer/ directory and move python files
• Fix paths in source code.

[Jeadie]

Example output

(.venv) jeadie@Jacks-MBP gpt-engineer % gpt-engineer --help
Usage: gpt-engineer [OPTIONS] [PROJECT_PATH]

Arguments:
  [PROJECT_PATH]  path

Options:
  --run-prefix TEXT               run prefix, if you want to run multiple
                                  variants of the same project and later
                                  compare them
  --model TEXT                    [default: gpt-4]
  --temperature FLOAT             [default: 0.1]
  --install-completion [bash|zsh|fish|powershell|pwsh]
                                  Install completion for the specified shell.
  --show-completion [bash|zsh|fish|powershell|pwsh]
                                  Show completion for the specified shell, to
                                  copy it or customize the installation.
  --help                          Show this message and exit.

[patillacode]

Hey @Jeadie

this looks great, although I left a couple of comments, only one requesting a change though.

couple of notes:

• can you make sure you get rid of all those old commits? I think you branched out in a weird state maybe?

• can you add the corresponding documentation to the README.md file?

I'll give this a go in the meantime.

EDIT:
I tried to run it, unsuccessfully.
It seems like running via the command doesn't work.

The error is this one:

FileNotFoundError: [Errno 2] No such file or directory: '/Users/dvitto/projects/gpt-engineer-jeadie-60/venv/lib/python3.11/site-packages/identity/qa'

Running with --help gave the expected output but
I believe it could be related to what @AntonOsika mentioned here.

Also, I just noticed, the parameters are not being taken care of.
Meaning passing parameters are not taken into account at all, am I missing something?

Thanks a lot.

[patillacode]

Good catch.

[patillacode]

I also like this better, explicit and probably saving someone from a circular import dependency.

[patillacode]

can't we just call app() instead of execute() and get rid of the execute definition?

[Jeadie]

This is helpful for referencing it in pyproject.toml

[Jeadie]

I think it will be easier (and cleaner/best practice going forward) to squash and merge all contributions to have a more readable main branch. Making some improvements in a second PR, will QA again today.

[Jeadie]

Fixes are up. They were just due to how i was testing things initially. Having identity as a file is highly fragile. We will need to move off it soon.

[patillacode]

It is now working, thanks for the changes.

Two more things:

• can you squash your commits?
• parameters (i,e --model) are not being taken care of, I believe you left that on purpose, is this something you plan on adding?

I don't feel comfortable merging something that tells the user "use me like this" (when running gpt-engineer --help) but it is not actually working unless we plan on adding it shortly.

[Jeadie]

I did not touch --model flag and --model flag seems to work on mine (passing an invalid value to --model invalid_id throws the correct error). We can squash commits at merge time, which is generally an easy approach.