The clear star of yesterday #2920
Conversation
alexdowad
force-pushed
the
the_clear_star_of_yesterday
branch
from
2376025
to
8186277
Compare
|
Apart from the build failure, the changes look good. |
alexdowad
force-pushed
the
the_clear_star_of_yesterday
branch
from
8186277
to
11646f2
Compare
|
Still have to fix up a style violation. Also profiled autocorrection of some very slow files from Rails. It turns out that For autocorrection to be fast, we need to use specific checks to detect cases where autocorrection should not be done. I have removed almost all uses of it. The only one left is |
|
Forgot to say: on one large test file from Action Pack which I was using for benchmarking, autocorrection is now more than 10x faster. |
| @@ -160,11 +160,11 @@ | |||
| end | |||
|
|
|||
| context 'in a method call without parentheses' do | |||
| it 'does not correct a hash parameter with trailing comma' do | |||
| it 'corrects a hash parameter with trailing comma' do | |||
| # Because `get :i, x: 1,` is invalid syntax. | |||
There was a problem hiding this comment.
This comment seems to suggest that your change produces invalid code?
There was a problem hiding this comment.
I'll delete the comment.
alexdowad
force-pushed
the
the_clear_star_of_yesterday
branch
from
843f45f
to
70da440
Compare
|
With this change, autocorrection of large and style-error-ridden projects like Rails is quite fast! My stress tests can be run comfortably. The very first run of the autocorrection stress test has picked up a bunch of bugs. Just looking at a problem where |
alexdowad
force-pushed
the
the_clear_star_of_yesterday
branch
from
70da440
to
c1706c8
Compare
|
Fixed up the broken build. |
alexdowad
force-pushed
the
the_clear_star_of_yesterday
branch
from
c1706c8
to
6ddff47
Compare
|
Rebased on master. |
|
Turns out the |
alexdowad
force-pushed
the
the_clear_star_of_yesterday
branch
from
6ddff47
to
f0a705a
Compare
|
One commit was removed, right? Please rebase on top of |
When the enforced style is `semantic`, `check_for(:raise, ...)` meant that we *don't* want to see `raise`. But with other styles, `check_for(:raise, ...)` meant that we *do* want to see `raise`. Make the usage consistent.
This helper had two different functions depending on the enforced style. Split it so each helper method does just one thing.
…stom `fail` methods
If a custom method called `fail` is defined anywhere in a source file, calls to
`fail` within that file are not flagged, even if the enforced style is
`only_raise`. This prevents unwanted offenses in situations like:
class Foo
def self.fail(msg)
# something
end
def self.foo
fail 'message'
end
end
…ue in braceless hash
Previously, the use of `{}` would be flagged in:
cr.stubs client: mock {
expects(:email_disabled=).with(true)
expects :save
}
However, changing this to use `do...end` would change the meaning of the code,
since it would bind to the outer method call rather than the inner one.
The changes pushed this class over the line count limit, so a couple of tweaks
have been made to save on excess code size.
There is a good reason for this; using AutocorrectUnlessChangingAST imposes an extremely high performance cost on autocorrection. Surprisingly, after removing it from AndOr, all the tests still pass. Likely, AndOr was not very smart before about which expressions to flag, so AutocorrectUnlessChangingAST was used to stop it from breaking code. Now, it may be smarter than before. It is also possible that there are some cases where it will perform invalid autocorrections. If that is the case, we can fix them when they are found (and add corresponding test cases).
Again, this provides a significant performance boost for autocorrection.
Another performance boost for autocorrection.
Finally, with the main users of AutocorrectUnlessChangingAST removed, autocorrection can really fly. One very slow file (actionpack/test/controller/routing_test.rb) from Rails, which I was using as a test case, is more than 10x faster now.
… autocorrecting A Ruby method works exactly the same when an unused block argument is removed. It also makes each call to the method faster.
alexdowad
force-pushed
the
the_clear_star_of_yesterday
branch
from
3bd12c5
to
e60ee2f
Compare
alexdowad
force-pushed
the
the_clear_star_of_yesterday
branch
2 times, most recently
from
23bc2f4
to
3d87cbf
Compare
It uses parentheses as necessary to avoid changing the meaning of an expression,
and better yet:
not a == b
...is corrected to:
a != b
It is also much faster when autocorrecting, since AutocorrectUnlessChangingAST
is not used any more.
It was convenient, but it killed performance.
alexdowad
force-pushed
the
the_clear_star_of_yesterday
branch
from
3d87cbf
to
7bd4ccd
Compare
|
And we're back to green. |
Before submitting a PR make sure the following are checked:
master(if not - rebase it)and description in grammatically correct, complete sentences.