Improve GT.save() usability

[jrycw]

Related PR: #496.

This PR introduces several improvements to enhance the usability of GT.save():

• Propose exposing **params to allow advanced users to customize save parameters directly via Pillow.
• Streamline the .png extension check to a single line.
• Construct the URL directly, similar to FmtImage._get_image_uri(), without using tempfile.TemporaryDirectory().
• Replace the final time.sleep(0.05) with WebDriverWait(driver, 1).until() (we may need to determine the optimal timeout for most use cases).
• Remove the .png saving branch that relied on selenium; all files are now saved using Pillow.
• Suggest modifying GT.save() to return itself, allowing users to save intermediate tables. For example:

  GT(exibble).save("initial.png").tab_header("title").save("with_title.png")

Additionally, I noticed a potential speed boost (approximately 40-50% faster on my machine) when calling headless_browser=wdriver(options=wd_options) directly, bypassing the context manager. However, I'm unsure about the safety implications of this approach.

[codecov]

Codecov Report

Attention: Patch coverage is 91.52542% with 5 lines in your changes missing coverage. Please review.

Project coverage is 89.38%. Comparing base (1ad3ec1) to head (23f12cb).
Report is 14 commits behind head on main.

Files with missing lines	Patch %	Lines
great_tables/_utils_selenium.py	90.90%	5 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #499      +/-   ##
==========================================
+ Coverage   87.86%   89.38%   +1.52%     
==========================================
  Files          42       45       +3     
  Lines        4852     5239     +387     
==========================================
+ Hits         4263     4683     +420     
+ Misses        589      556      -33     

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

---

🚨 Try these New Features:

• Flaky Tests Detection - Detect and resolve failed and flaky tests

[jrycw]

Hello team,

In my last commit, I experimented with isolating the web driver preparation logic into _utils_selenium.py. Please feel free to review and decide whether to keep or discard this change.

[machow]

Could we consolidate the driver initialization across all these subclasses, by adding cls_wd_options=ChromeOptions, and cls_driver=Chrome as class attributes?

That way, the BaseClass could essentially require subclasses to specify those pieces. Then, the init would not have to be overridden by child classes.

This makes the child classes mostly about:

• specifying Driver and Option classes (as class attributes)
• customizing add_arguments method

[jrycw]

@machow , thanks for the quick response.

Using cls_driver seems like a good idea, but I’m concerned that cls_wd_options might lead to issues. For example, what if we need to create two Chrome instances with different ChromeOptions in the future?

[jrycw]

@machow , I think I understand now. Does the latest commit align with what you had in mind?

[machow]

Yeah, thanks -- that looks really nice!

[rich-iannone]

This is looking good! I had a look at the output images on all 4 webdrivers (at 3 different scale values) to make sure outputs are similar to main. In case this is interesting I've included the images in this zip file:

image-tests.zip

I've found no reduction in quality with the changes here. Chrome is the only one that is problematic (cuts off content at the bottom) but that's also on main and an open issue anyway.

[machow]

This PR is so helpful! Thanks for taking the time to work on .save(), which afaict is one of the most important parts of our API. My one request was that we remove **params so that PIL is a internal detail.

Excited to merge all these PRs, sorry for slowing your roll 😭

[machow]

Is it okay if we remove the **params piece of this PR? I'm onboard with everything else. Because **params only applies to when we go from a .png -> something else, it exposes PIL as part of the user API.

If we leave it out, we can swap out PIL down the road. As an alternative could we tell users that we're using Pillow? We could even tell them about PIL open and save if needed (so they can go from .png -> anything PIL supports on their own if more customization is needed).

[jrycw]

Thanks for the explanation. I agree that we should remove **params, making Pillow an internal dependency rather than a public one.

Regarding hints, I believe that diligent readers likely already recognize we use Pillow under the hood and would consult its documentation for further customization if needed.

This aligns with our goal of providing a user-friendly way to save generated tables rather than focusing on creating highly customized table figures in various formats🤔.

[jrycw]

This is looking good! I had a look at the output images on all 4 webdrivers (at 3 different scale values) to make sure outputs are similar to main. In case this is interesting I've included the images in this zip file:

image-tests.zip

I've found no reduction in quality with the changes here. Chrome is the only one that is problematic (cuts off content at the bottom) but that's also on main and an open issue anyway.

As a side note, this PR does not address the cutoff bug encountered when using Chrome, as mentioned in #480 .

[rich-iannone]

LGTM!

[machow]

Shoot, I just noticed the change to passing a base64 encoded url directly to the headless browser. I think there could be some challenges with this approach (and url length limitations). In general, modern browsers seem to have large limits, but it's a tricky territory.

Opened an issue to track:

#518