Considering an alternative to pynliner about rssant HOT 4 CLOSED

Hi @Stranger6667, thank you for the proposal!

About Performance:

pynliner is really slow and I have ever encountered performance issue when render templates/email/feed_analysis.html.mako. The template contains a very large table, and need inline css for each cell. So the render process is very slow, and the final html is very large, which also cause email app rendering very slow.

My solution is pre inline css on mako template and cache the template, then do mako rendering. So only need inline css once, and reuse the inlined template. FYI: email_template.py

And to avoid too large html, I have to use css classes instead of inline styles.

it might require changing the order of rendering for mako templates (pynliner doesn't transform Mako's syntax as css-inline does)

It's not suitable for my cases to change rendering order, as described above.

About ARM wheel

ARM64 is required for rssant. Not sure whether it's hard to install from source or not. It's possible to build ARM64 wheel from source via docker multi-stage build.

from rssant.

@guyskk Thank you for providing the context!

Re ARM64 wheel: I added wheels for Python 3.8+ on macOS ARM64 and some more platforms including manylinux aarch64, musllinux aarch64 etc. Hopefully, it satisfies the requirements :) Let me know if it is not sufficient, and rssant needs e.g. Python 3.7 on ARM64

Re performance: Pre-inlining is a neat trick indeed! :) I tried to render that template with some randomly generated data (see below) and post-inlining with pynliner takes ~2.3s vs ~8.2ms with css-inline. Though, with pre-inlining, it becomes much or less irrelevant, as switching to post-inlining will give a net-negative effect on performance in any case :) Probably it is something for me to consider - maybe css-inline may support templated input one day :)

from hypothesis import provisional as pr, strategies as st, settings, given

...  # import `Feed` from feed_analysis here - locally I copied the relevant definitions

feeds = st.builds(
    Feed, 
    id=st.integers(min_value=1), 
    domain=pr.domains(), 
    status=st.sampled_from(["pending", "updating", "ready", "error", "discard"]), 
    use_proxy=st.booleans(), 
    response_status=st.integers(min_value=200, max_value=500), 
    freeze_level=st.integers(min_value=1, max_value=10)
)

out = []

@given(feeds)
@settings(max_examples=10000)
def generate(x):
     out.append(x)

# calling generate() will add 10000 random `Feed` instances to `out`
# Then generating snapshots from it via `compute_snapshot`, and passing them to `compute_report`

So, effectively it looks like the only way to use css-inline at the moment means a performance hit due to unconditional post-inlining (though, maybe template itself can be re-arranged to support css-inline, but it is too much for this change). premailer also malformed the template :( Let me know if there is anything I can do on my side here, and thank you for consideration :)

from rssant.

@Stranger6667 Seems nothing needs to be done for inlining css in rssant project.

If you have interest in high performance project, maybe take a look at this optics science project:

• opticspy/lightpipes#57
• opticspy/lightpipes#43

from rssant.

Thanks, @guyskk ! :)

from rssant.