Deprecate direct access to globals like debug and verbose.

[obi1kenobi]

Instead of accessing langchain.debug, langchain.verbose, or langchain.llm_cache, please use the new getter/setter functions in langchain.globals:

• langchain.globals.set_debug() and langchain.globals.get_debug()
• langchain.globals.set_verbose() and langchain.globals.get_verbose()
• langchain.globals.set_llm_cache() and langchain.globals.get_llm_cache()

Using the old globals directly will now raise a warning.

[vercel]

The latest updates on your projects. Learn more about Vercel for Git ↗︎

Name	Status	Preview	Comments	Updated (UTC)
langchain	✅ Ready (Inspect)	Visit Preview	💬 Add feedback	Oct 12, 2023 10:47pm

[obi1kenobi]

Still needs:

• better deprecation warning message that points users to the new code to use
• conversion of internal write uses to the new functions
• conversion of internal read uses to make them use the new code path instead of using top-level langchain import
  • There's an unfortunate Python quirk: the __getattr__() in langchain/__init__.py creates a separate copy of the verbose and debug variables, so setting langchain.verbose with this PR does not actually set langchain.utils.globals._verbose.
  • I implemented a workaround: the new explicit setter functions set both copies, and the explicit getters will read both copies and return whichever one was set. That way using the new getters and setters is backward-compatible, and uses of the old top-level langchain values will also be able to set the values and see any changes done to them via the new APIs as well.
  • Using the old langchain top-level values directly now triggers a warning with a concrete suggestion to use the corresponding explicit get/set functions. At some point in the future when users have migrated to using the new API, we can remove the top-level values entirely.

[hwchase17]

seems reasonable, lets update all the docs (and internal usage) that do langchain.verbose=True to use these

[obi1kenobi]

I also tried this sort of thing and hoped it would work, but no luck.

PEP 562 which defines the module-level __getattr__() doesn't allow for a __setattr__() so I don't think this ever gets called: https://peps.python.org/pep-0562/

[hwchase17]

ah whoops i tried this out, it didnt work, and i forgot to remove 🤦

[obi1kenobi]

Thanks for updating the docs, I completely forgot they are in the same repo. Will keep in mind for the future.

[xb3sox]

Any idea how to remove this warning, it appear after each message:

D:\Agents\langchain\.venv\Lib\site-packages\langchain\__init__.py:34: UserWarning: Importing llm_cache from langchain root module is no longer supported. Please use langchain.globals.set_llm_cache() / langchain.globals.get_llm_cache() instead.
  warnings.warn(

[obi1kenobi]

Any idea how to remove this warning, it appear after each message:

D:\Agents\langchain\.venv\Lib\site-packages\langchain\__init__.py:34: UserWarning: Importing llm_cache from langchain root module is no longer supported. Please use langchain.globals.set_llm_cache() / langchain.globals.get_llm_cache() instead.
  warnings.warn(

I believe this might have been fixed in a subsequent PR, though I'm not exactly sure if it was already released or not yet. Please try upgrading langchain, and if the problem persists consider opening a new issue for it.