chrome/chromium renderer not detected correctly on Ubuntu (name is chromium-browser)

[emmanuelle]

On my Ubuntu machine I cannot configure a chrome renderer (my default browser is firefox). The reason is that Chromium is registered with the name chromium-browser as shown below, so that webbrowser.get throws an error when called with chrome.

>>> import plotly.io as pio                                                                                                      
>>> pio.renderers.default = 'chromium'                                                                                           
>>> import plotly.graph_objects as go                                                                                            
>>> go.Figure(go.Scatter(x=[1], y=[0])).show()                                                                                   
Traceback (most recent call last):
  File "<ipython-input-5-d0da89a1679b>", line 1, in <module>
    go.Figure(go.Scatter(x=[1], y=[0])).show()
  File "/home/emma/code/plotly.py/packages/python/plotly/plotly/basedatatypes.py", line 2794, in show
    return pio.show(self, *args, **kwargs)
  File "/home/emma/code/plotly.py/packages/python/plotly/plotly/io/_renderers.py", line 391, in show
    renderers._perform_external_rendering(fig_dict, renderers_string=renderer, **kwargs)
  File "/home/emma/code/plotly.py/packages/python/plotly/plotly/io/_renderers.py", line 340, in _perform_external_rendering
    renderer.render(fig_dict)
  File "/home/emma/code/plotly.py/packages/python/plotly/plotly/io/_base_renderers.py", line 736, in render
    open_html_in_browser(html, self.using, self.new, self.autoraise)
  File "/home/emma/code/plotly.py/packages/python/plotly/plotly/io/_base_renderers.py", line 681, in open_html_in_browser
    webbrowser.get(using).open(
  File "/usr/lib/python3.7/webbrowser.py", line 65, in get
    raise Error("could not locate runnable browser")
Error: could not locate runnable browser

>>> import webbrowser                                                                                                            
>>> print(webbrowser._browsers)                                                                                                  
{'xdg-open': [None, <webbrowser.BackgroundBrowser object at 0x7fe81203fef0>], 'gvfs-open': [None, <webbrowser.BackgroundBrowser object at 0x7fe81203fbe0>], 'x-www-browser': [None, <webbrowser.BackgroundBrowser object at 0x7fe81203f7b8>], 'firefox': [None, <webbrowser.Mozilla object at 0x7fe81203ff98>], 'chromium-browser': [None, <webbrowser.Chrome object at 0x7fe81203f860>]}

We could either add another renderer name chromium-browser, or query matching strings in webbrowser._browser.keys(). The second solution is probably cleaner and could also make it possible to remove either 'chrome' or 'chromium' if we want. Thoughts @jonmmease @nicolaskruchten ?

[nicolaskruchten]

Happy to go with the cleaner solution. What does setting the renderer to browser do?

[emmanuelle]

It uses the default browser.

[emmanuelle]

I wanted to reproduce a problem which a user was having on Chrome without having to change my default browser.

[jonmmease]

... or query matching strings in webbrowser._browser.keys()

Yeah, I like this idea. I think attempting to separate chrome and chromium is still a good idea. We could have the BrowserRenderer class accept a list of strings as the using argument, and have it use the first browser in the list that's present in webbrowser._browser.keys. Then we could have lines like

renderers["chrome"] = BrowserRenderer(config=config, using=["chrome", "chrome-browser"])
renderers["chromium"] = BrowserRenderer(config=config, using=["chromium", "chromium-browser"])

[c-chaitanya]

Hi @emmanuelle @jonmmease @nicolaskruchten

I have gone through the code and realised that this issue isn't fixed yet. Just wanted to confirm if I could start working on the fix or if it was being worked on by someone else?

[nicolaskruchten]

A fix would be welcome, thank you :)

[c-chaitanya]

Thanks for the response @nicolaskruchten

Running webbrowser._browser.keys gives me the following list on my Ubuntu 16.04 distro

['firefox', 'xdg-open', 'gvfs-open', 'x-www-browser', 'google-chrome', 'chromium-browser', 'opera']

From @emmanuelle code we see that using chromium (in line 2) breaks the code , simillarly using chrome also breaks the code on my system.
Again using OS registered strings like chromium-browser or google-chrome does not throw any error.

import plotly.io as pio                                                                                                      
pio.renderers.default = 'chromium'                                                                                           
import plotly.graph_objects as go                                                                                            
go.Figure(go.Scatter(x=[1], y=[0])).show()

Experimenting with string matching seems to be the best solution(as was discussed above).
I could come up with this code for the same

self.using = using
    # Checks if self.using is not None and that it isn't available in the list returned by webbrowser._browsers.keys()
    if self.using and self.using not in webbrowser._browsers.keys():
       # Performs substring matching to fetch the browser closest to the provided name
       self.using = [browser for browser in webbrowser._browsers.keys() if self.using in browser][0]

let me know your views on the same.
Thank you

[jonmmease]

I think you've got the right idea. My intuition here would be to allow the using argument to open_html_in_browser to be a tuple of strings.

plotly.py/packages/python/plotly/plotly/io/_base_renderers.py

Line 652 in b025da3

def open_html_in_browser(html, using=None, new=0, autoraise=True):

Like I mentioned in #2348 (comment), then open_html_in_browser could scan through webbrowser._browser.keys and use the first entry that matches the using argument (Either an exact match if a string, or any entry in the tuple).

For import time, we want to make sure that none of this logic happens on import. So it shouldn't happen in BrowserRenderer the constructor.

[c-chaitanya]

Thank you for the suggestion @jonmmease. I have made the changes you have suggested in my local branch ( c-chaitanya@5ae493c ).

[c-chaitanya]

Hi @jonmmease, understanding other parts of the renderer code more clearly, I see that renderers are also invoked as follows chrome+colab+notebook+vscode, there is a method to also check if isinstance is a string such that they could be split on +.

If we have a look at the changes in these two files ( c-chaitanya@5ae493c ) we see that the parameters passed to using
in renderers["chrome"] = BrowserRenderer(config=config, using=("chrome", "google-chrome")) makes no difference even if it a string as follows renderers["chrome"] = BrowserRenderer(config=config, using="chrome"), it would internally perform string matching to fetch the closest matching browser. If using is a tuple it breaks integration with other renderer types.

[jonmmease]

Hi @c-chaitanya. Your changes in that commit look fine to me, I don't quite follow what issue you're running into.

The chrome in "chrome+colab+notebook+vscode" refers to the name of the renderer (renderers["chrome"]) and doesn't have any relationship with the value passed to the using argument of BrowserRenderer. So I don't think there's any problem here, but let me know if you're running into a problem.

Also, feel free to go ahead and make a PR even if you're not sure it's finished yet. It'll be a little easier to talk through specific code questions that way. Thanks!

[c-chaitanya]

Hi @jonmmease , I have given a pull request #3278 to fix this issue. The CircleCi fails for one test "Check formating with black". The thing was when I ran black on my local branch it gave me "903 files reformatted, 10053 files left unchanged". 903 was a big number so i did not add the reformatted files. Shoud I reformat and give a new pull request?

[c-chaitanya]

Hi @jonmmease @nicolaskruchten I reformatted using black and gave a pull request again. This time I notice CircleCI fails for test related to orca. Just to check if errors were due to my changes, I ran the pytests on the plotly.py master branch and I notice the same orca errors. Thus I deduced that the errors preexisted in the master branch and were not induced by these changes. Is there anything I can do to fix these errors?.

[nicolaskruchten]

Closed in #3278 and improved in #3310

[nicolaskruchten]

Thanks @c-chaitanya 👏