Perf: defer heavy startup imports to reduce launch time by ~500ms

[eendebakpt]

Problem

Spyder's startup time was unnecessarily slow due to several third-party libraries being imported eagerly at module level, even though they are only needed lazily or in rare code paths.

Profiled using python -X importtime and subprocess timing.

Changes

spyder/utils/programs.py — removes requests and defers asyncio

• requests (~250ms saved): CaseInsensitiveDict from requests.structures was imported at module level but only used in one Windows-only branch inside alter_subprocess_kwargs_by_platform(). Replaced with an inline case-insensitive key lookup using only stdlib ({k.lower(): k for k in env}).
• asyncio (~100ms saved, deferred): Moved inside the if asynchronous: branch of run_shell_command(), which is called only from environ.py with asynchronous=True.

spyder/config/manager.py — defers keyring (~200ms saved)

keyring and keyring.errors.NoKeyringError were imported at the top of the module. These are only needed inside if secure: branches (get, set, remove_option), which are only hit when remote/credential-secured config options are accessed. Moved the imports to those three call sites.

spyder/config/appearance.py — avoids importing sphinxify (~300ms saved)

from spyder.plugins.help.utils.sphinxify import CSS_PATH was triggering the full sphinx + docutils + jinja2 import chain at startup, just to get a path string constant. Replaced with importlib.util.find_spec('spyder.plugins.help.utils') which resolves the path without loading the Sphinx stack.

Measured improvement

Module	Before	After
spyder.utils.programs	~440ms	~130ms
spyder.config.manager	~616ms	~220ms
spyder.app.start	~431ms	~167ms

🤖 Generated with Claude Code

[ccordoba12]

Hey @eendebakpt, why did you close this PR? The gains in startup time seem significant enough to merit its inclusion in our next version.

[eendebakpt]

@ccordoba12 PR was opened too soon by Claude. I wanted to review the PR myself first to avoid wasting reviewers resources and I wanted to validate the actual gains. Now that you are looking anyway: if change makes sense just reopen it and we can iterate until the PR is good.

For python 3.15 I would recommend to adopt PEP 810 (it will help with more imports).

[ccordoba12]

@dalthviz, please take a look at @eendebakpt's work.

[dalthviz]

I think here we could use something like what is done over the IPython Console to prevent having to do a import spyder:

spyder/spyder/plugins/ipythonconsole/widgets/client.py

Lines 73 to 75 in 127e840

	PLUGINS_PATH = get_module_source_path('spyder', 'plugins')
	CSS_PATH = osp.join(PLUGINS_PATH, 'help', 'utils', 'static', 'css')

However, maybe it could be worthy to move the CSS_PATH constant definition somewhere from where every other module could import it and without causing other unwanted modules to get imported? 🤔

Also. as a side note, checking the CSS_PATH constant usage seems like we have a TODO related with it around the mainwindow setup method (import that maybe could be affecting startup times too?):

spyder/spyder/app/mainwindow.py

Lines 659 to 662 in 127e840

	# Set css_path as a configuration to be used by the plugins.
	# TODO: Remove circular dependency between help and ipython console
	# and remove this import. Help plugin should take care of it
	from spyder.plugins.help.utils.sphinxify import CSS_PATH, DARK_CSS_PATH

What do you think @ccordoba12 ?

[eendebakpt]

@dalthviz Do you want me to look into this?

[dalthviz]

/show binder

[github-actions]

👈 Launch a Binder instance on this branch