Handle use case when canvas size is already controlled by css

[ypujante]

When I was checking the SDL implementation I noticed that they were doing something like this:

    emscripten_set_canvas_element_size(wdata->canvas_id, 1, 1);
    emscripten_get_element_css_size(wdata->canvas_id, &css_w, &css_h);
    wdata->external_size = SDL_floor(css_w) != 1 || SDL_floor(css_h) != 1;

Then they use this flag wdata->external_size to not rewrite the css property, for example:

            if (!window_data->external_size && window_data->pixel_ratio != 1.0f) {
                emscripten_set_element_css_size(window_data->canvas_id, w, h);
            }

I changed the glfw code to use the same "trick" and implement a similar logic. This way there is no need to wrap the canvas in a <div> when the canvas needs to be full window (width: 100% / height: 100%)

[sbc100]

Thanks!

lgtm % nits

[ypujante]

@sbc100 I applied your code review. Thank you.

[ypujante]

All tests ran.

The primary test for this change is working fine:

  test_glfw3_hi_dpi_aware (test_browser.browser) ... cache:INFO: generating system asset: symbol_lists/0bef552e4ea38aecbf95a4972a5e9b306b78bc15.json... (this will be cached in "/root/cache/symbol_lists/0bef552e4ea38aecbf95a4972a5e9b306b78bc15.json" for subsequent builds)
cache:INFO:  - ok
ok (1.179s)

The only failing test in test-browser-firefox is

FAIL [121.660s]: test_wasmfs_fetch_backend_bigint (test_browser.browser)

  test_wasmfs_fetch_backend_bigint (test_browser.browser) ... cache:INFO: generating system asset: symbol_lists/1d95e008792744511b35b1f1168933b8b40d1d9e.json... (this will be cached in "/root/cache/symbol_lists/1d95e008792744511b35b1f1168933b8b40d1d9e.json" for subsequent builds)
cache:INFO:  - ok
[unresponsive tests: 1]
common:INFO: Restarting browser process
Exiting due to channel error.
Exiting due to channel error.
Exiting due to channel error.
Exiting due to channel error.
Exiting due to channel error.
Exiting due to channel error.
Exiting due to channel error.
common:INFO: Launching browser: ['/root/firefox/firefox', '-headless', '-profile', '/root/tmp-firefox-profile/']
[test error (see below), automatically retrying]
Expected to find '/report_result?exit:0
' in '[no http server activity]
', diff:

--- expected
+++ actual
@@ -1 +1 @@
-/report_result?exit:0
+[no http server activity]

This failing test does not seem to be related to my changes (note that test_glfw3_hi_dpi_aware is skipped for the suite test-browser-firefox)

Thank you

[ypujante]

@sbc100 I just pushed a small addition which does 2 things:

• adds a test for the new feature (use case 8), making sure that if the canvas has a css override, then it is doing what the code change is supposed to be doing (meaning the override takes precedence)
• store/restore the original canvas size to be a good citizen. This is not really related to my changes, and this glfw implementation has never done it before, but I figured it made sense to restore the canvas the way it was before glfwCreateWindow is called

[sbc100]

lgtm!