internet.browsers.engines.gosub
How does an HTML5 parser handle broken markup?
The spec's error recovery rules seem enormous. How do engines keep that manageable?
internet.browsers.engines.gosub
The spec's error recovery rules seem enormous. How do engines keep that manageable?
Read the spec with a real page open next to it. Abstract rules make more sense with an example.
Reference screenshots plus a fuzzy pixel diff catch most regressions.
Honestly the cheap option was fine for the first few years.
Layout bugs often turn out to be cascade bugs in disguise.
Agreed, and I'd add that it takes a while before it clicks.
I'd be careful with that advice, it worked for me but I know people it didn't work for.
I tried this last year and ended up somewhere completely different. Not worse, just different.
misty-rowan-88 wrote:
I'd be careful with that advice, it worked for me but I know people it didn't work for.
Same experience here.
The second time went a lot better once I stopped rushing it.
Strong disagree, but I appreciate the write-up.
Honestly the cheap option was fine for the first few years.
Same experience here.
The second time went a lot better once I stopped rushing it.
Start with the CSS parser. It's self-contained and well specified.
I'd be careful with that advice, it worked for me but I know people it didn't work for.
still-willow-23 wrote:
Start with the CSS parser.
I'd be careful with that advice, it worked for me but I know people it didn't work for.
late-willow-50 wrote:
still-willow-23 wrote:
That's fair. I hadn't looked at it that way.
gentle-lichen-54 wrote:
late-willow-50 wrote:
Could you share a photo of how it turned out?
The web platform tests are the best map of what's still missing.
Agreed, and I'd add that it takes a while before it clicks.
Follow the tokenizer and tree builder state machines literally. Clever shortcuts break on real pages.
Do you have a source for that? Not doubting, I just want to read more.
Strong disagree, but I appreciate the write-up.
amber-sedge-23 wrote:
Follow the tokenizer and tree builder state machines literally.
Honestly the cheap option was fine for the first few years.
Do you have a source for that? Not doubting, I just want to read more.
Implement the tokenizer and tree construction states exactly as written, shortcuts fail on real-world HTML.
I tried this last year and ended up somewhere completely different. Not worse, just different.