Faster feature checks #17229

tonycoz · 2019-10-28T20:06:11Z

No description provided.

atoomic · 2019-10-28T20:12:40Z

this excellent!
maybe once merged we could consider using back the original syntax
s{FEATURE_IS_ENABLED_MASK}{FEATURE_IS_ENABLED}g ?

atoomic · 2019-10-28T20:19:25Z

toke.c

- * Check whether the named feature is enabled.
- */
-bool
-Perl_feature_is_enabled(pTHX_ const char *const name, STRLEN namelen)


note that we will also have to remove or adjust feature_is_enabled from Devel::PPPort

Probably remove, feature_is_enabled was only used by FEATURE_IS_ENABLED(), which is only intended for use by core or included extensions (PERL_EXT).

leonerd · 2019-10-28T22:07:57Z

regen/feature.pl

@@ -76,6 +76,17 @@ BEGIN
 ###########################################################################
 # More data generated from the above

+if (keys %feature > 32) {
+    die "cop_features only has room for 32 features";


What would be the thing to do if we run out of bits? Can we easily turn the field into a U64, or would we make a second one?

If U64 can be used, it may make sense to reuse cop_hints (it's a U32, but 28 of its bits are currently in use)

Simplest for working with feature.pm/${^FEATURE_BITS} would probably be a U8 array, but that's difficult to write a SAVEFEATUREBITS() for.

We could make it a pointer as for warnings, but since the feature bits are fixed size and will be present in any cop, the pointer is just overhead.

But for now we have over twice the feature bits as features, so I don't feel much urgency to decide right now.

If U64 can be used, it may make sense to reuse cop_hints (it's a U32, but 28 of its bits are currently in use)

If we worked through $^H we'd need to require -Duse64bitint to safely update it from feature.pm

leonerd

LGTM

Leont · 2019-10-28T23:24:07Z

${^FEATURE_BITS} should probably be documented in perlvar

Perform only a bit check instead of a much more expensive hash lookup to test features. For now I've just added a U32 to the cop structure to store the bits, if we need more we could either add more bits directly, or make it a pointer. We don't have the immediate need for a pointer that warning do since we don't dynamically add new features during compilation/runtime. The changes to %^H are retained so that caller() can be used from perl code to check the features enabled at a given caller's scope.

haarg · 2019-10-29T15:07:56Z

This currently leaves no way for a user to manually construct a ${^FEATURE_BITS} value. While the bits themselves are internal values, it would still be nice to be able to construct a value based on feature names.

leonerd · 2019-10-29T15:27:09Z

@haarg Perhaps this could be solved by feature.pm adding a function?

$bits = feature::make_bitmask(qw( say evalbytes ));

haarg · 2019-10-29T15:32:00Z

That would make sense to me.

tonycoz · 2019-10-30T00:16:48Z

This currently leaves no way for a user to manually construct a ${^FEATURE_BITS} value. While the bits themselves are internal values, it would still be nice to be able to construct a value based on feature names.

Why do they need it?

If they want to enable features they can use feature->(un)import() in the current compilation.

It also risks the feature bits and %^H feature entries getting out of sync.

I'd rather remove ${^FEATURE_BITS} rather than even vaguely suggest it's something users should be touching.

That said, there is a way to get the bits for a given feature set: call feature->import() at runtime with the features you want, and extract the bits.

leonerd · 2019-10-30T13:57:27Z

It remains to be seen whether such a feature would be needed, but it can be added in a small addition to feature.pm later if required. Its absence now shouldn't hold back this particular change, which basically just improves performance to no real detriment in API. I believe it should be merged.

tonycoz · 2019-10-30T19:10:26Z

I've merged this, I may end up removing ${^FEATURE_BITS} (but keeping the optimization)

Leont · 2019-10-30T23:37:42Z

unicode_strings is now living in both the hints and the features bitsets, this doesn't seem necessary.

tonycoz · 2019-11-03T21:07:11Z

unicode_strings is now living in both the hints and the features bitsets, this doesn't seem necessary.

I had a look at this, I don't think it's worth reinstating FEATURE_IS_ENABLED() / Perl_feature_is_enabled() to save this bit. Opinions welcome.

Leont · 2019-11-04T19:05:40Z

It's not saving a bit that I'm worried about, but more things like "what happens if they're out of sync?". Also, feature.pm now contains special case code for unicode_strings that could be eliminated if that feature no longer uses the hints bitset.

tonycoz · 2019-11-04T21:56:01Z

The out of sync could happen between $^H{feature_unicode} and the bit in PL_hints before this.

Losing sync between $^H{feature_*} and ${^FEATURE_BITS} seems more likely if the user modifies either of those outside of feature.pm, but I see that as a "Doctor, it hurts when I do this" type of issue.

I've considered eliminating the visibility of ${^FEATURE_BITS} and adjusting the magic in Perl_magic_(set|clear)hints() to update cop_features instead, but the perl code is easier to maintain.

cpansprout · 2019-11-29T06:49:58Z

There are modules that copy the values of $^H, %^H, and ${^WARNING_BITS} from one lexical scope to another. So for %^H to get out of synch with ${^FEATURE_BITS} is not unlikely to happen.

I think having yet another lexical hints variable is something that should be avoided.

atoomic reviewed Oct 28, 2019

View reviewed changes

atoomic approved these changes Oct 28, 2019

View reviewed changes

leonerd reviewed Oct 28, 2019

View reviewed changes

leonerd approved these changes Oct 28, 2019

View reviewed changes

tonycoz added 3 commits October 29, 2019 10:32

remove now unneeded FEATURE_IS_ENABLED()

db39fc6

add ${^FEATURE_BITS} to perlvar.pod

298a079

tonycoz force-pushed the faster-feature branch from 712b4ac to 298a079 Compare October 28, 2019 23:45

leonerd mentioned this pull request Oct 30, 2019

Add an isa operator #17212

Closed

1 task

tonycoz merged commit df31f96 into Perl:blead Oct 30, 2019

tonycoz deleted the faster-feature branch November 4, 2019 21:40

Faster feature checks #17229

Faster feature checks #17229

Uh oh!

Conversation

tonycoz commented Oct 28, 2019

Uh oh!

atoomic commented Oct 28, 2019

Uh oh!

atoomic Oct 28, 2019 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

tonycoz Oct 28, 2019

Choose a reason for hiding this comment

Uh oh!

leonerd Oct 28, 2019

Choose a reason for hiding this comment

Uh oh!

Leont Oct 28, 2019 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

tonycoz Oct 28, 2019

Choose a reason for hiding this comment

Uh oh!

tonycoz Oct 28, 2019

Choose a reason for hiding this comment

Uh oh!

leonerd left a comment

Choose a reason for hiding this comment

Uh oh!

Leont commented Oct 28, 2019

Uh oh!

haarg commented Oct 29, 2019

Uh oh!

leonerd commented Oct 29, 2019

Uh oh!

haarg commented Oct 29, 2019

Uh oh!

tonycoz commented Oct 30, 2019 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

leonerd commented Oct 30, 2019

Uh oh!

tonycoz commented Oct 30, 2019

Uh oh!

Leont commented Oct 30, 2019

Uh oh!

tonycoz commented Nov 3, 2019

Uh oh!

Leont commented Nov 4, 2019

Uh oh!

tonycoz commented Nov 4, 2019

Uh oh!

cpansprout commented Nov 29, 2019

Uh oh!

Uh oh!

atoomic Oct 28, 2019 •

edited

Loading

Leont Oct 28, 2019 •

edited

Loading

tonycoz commented Oct 30, 2019 •

edited

Loading