-
-
Save cjepeway/42215154aff709d3efd7 to your computer and use it in GitHub Desktop.
#!/usr/bin/env perl6 | |
constant RAKUDO-SRC = '/home/jepeway/src/tz/rakudo-fork'; | |
#say "[working]"; | |
for lines() { | |
#say "checking [$_]"; | |
.say; | |
next unless m{(<[ . , ; \w \d / \- + @ ]>*) ':' (.*)}; | |
my ($file, $lineno) = ($0, $1); | |
$file = RAKUDO-SRC ~ '/' ~ $file | |
unless $file.IO.e; | |
next unless $file.IO.e; | |
#say "$file:$lineno"; | |
my $fh = open $file; | |
my $n = 0; | |
for $fh.lines() { | |
next unless ++$n == $lineno; | |
"• $_".say; | |
last | |
} | |
$fh.close(); | |
} |
jepeway@go-go-go:~/src/tz/p6$ make t/05-datetimex P6=/home/jepeway/src/tz/rakudo-fork/install/bin/perl6 | |
prove --nocolor -e '/home/jepeway/src/tz/rakudo-fork/install/bin/perl6 -I /home/jepeway/src/tz/p6' -v t/05-datetimex | |
t/05-datetimex .. | |
1..8 | |
ok 1 - UTC - seconds since posix epoch | |
not ok 2 - American/New_York - seconds since posix epoch | |
# Failed test 'American/New_York - seconds since posix epoch' | |
# at t/05-datetimex line 27 | |
# expected: '0' | |
# got: '3600' | |
abstract method unimplemented in subclass | |
in method utc-offset-in-seconds at /home/jepeway/src/tz/p6/TimeZone.pm:6 | |
in method posix at /home/jepeway/src/tz/p6/DateTimeX.pm:7 | |
in method in-timezone at /home/jepeway/src/tz/p6/DateTimeX.pm:42 | |
in method new at src/gen/m-CORE.setting:20304 | |
in method new at /home/jepeway/src/tz/p6/DateTimeX.pm:34 | |
in method later at src/gen/m-CORE.setting:20399 | |
in method earlier at src/gen/m-CORE.setting:20451 | |
in block <unit> at t/05-datetimex:39 | |
# Looks like you planned 8 tests, but ran 2 | |
# Looks like you failed 1 tests of 2 | |
Dubious, test returned 1 (wstat 256, 0x100) | |
Failed 7/8 subtests | |
Test Summary Report | |
------------------- | |
t/05-datetimex (Wstat: 256 Tests: 2 Failed: 1) | |
Failed test: 2 | |
Non-zero exit status: 1 | |
Parse errors: Bad plan. You planned 8 tests but ran 2. | |
Files=1, Tests=2, 0 wallclock secs ( 0.01 usr 0.00 sys + 0.57 cusr 0.00 csys = 0.58 CPU) | |
Result: FAIL | |
make: *** [t/05-datetimex] Error 1 |
jepeway@go-go-go:~/src/tz/p6$ bt < /tmp/z | |
jepeway@go-go-go:~/src/tz/p6$ make t/05-datetimex P6=/home/jepeway/src/tz/rakudo-fork/install/bin/perl6 | |
prove --nocolor -e '/home/jepeway/src/tz/rakudo-fork/install/bin/perl6 -I /home/jepeway/src/tz/p6' -v t/05-datetimex | |
t/05-datetimex .. | |
1..8 | |
ok 1 - UTC - seconds since posix epoch | |
not ok 2 - American/New_York - seconds since posix epoch | |
# Failed test 'American/New_York - seconds since posix epoch' | |
# at t/05-datetimex line 27 | |
# expected: '0' | |
# got: '3600' | |
abstract method unimplemented in subclass | |
in method utc-offset-in-seconds at /home/jepeway/src/tz/p6/TimeZone.pm:6 | |
• multi method utc-offset-in-seconds(DateTime $when) returns Int { self.unimplemented(&?ROUTINE); } | |
in method posix at /home/jepeway/src/tz/p6/DateTimeX.pm:7 | |
• method posix() { | |
in method in-timezone at /home/jepeway/src/tz/p6/DateTimeX.pm:42 | |
• my DateTime $dt .= new(self.posix, :timezone($tz.utc-offset-in-seconds(self))); | |
in method new at src/gen/m-CORE.setting:20304 | |
• self.bless(:$year, :$month, :$day, | |
in method new at /home/jepeway/src/tz/p6/DateTimeX.pm:34 | |
• my $dt = self.new: floor($p - $leap-second).Int, :&formatter, :$timezone; | |
in method later at src/gen/m-CORE.setting:20399 | |
• return self.new(self.Instant + $amount, :timezone($.timezone)); | |
in method earlier at src/gen/m-CORE.setting:20451 | |
• method earlier(*%unit) { | |
in block <unit> at t/05-datetimex:39 | |
• my $time-change-utc = DateTimeX.new(:year(2014), :month(3), :day(9), :hour(2), :minute(0), :second(0), :timezone(UTC-TZ.new())).earlier(seconds => $off); | |
# Looks like you planned 8 tests, but ran 2 | |
# Looks like you failed 1 tests of 2 | |
Dubious, test returned 1 (wstat 256, 0x100) | |
Failed 7/8 subtests | |
Test Summary Report | |
------------------- | |
t/05-datetimex (Wstat: 256 Tests: 2 Failed: 1) | |
Failed test: 2 | |
Non-zero exit status: 1 | |
Parse errors: Bad plan. You planned 8 tests but ran 2. | |
Files=1, Tests=2, 0 wallclock secs ( 0.01 usr 0.00 sys + 0.57 cusr 0.00 csys = 0.58 CPU) | |
Result: FAIL | |
make: *** [t/05-datetimex] Error 1 |
And it seems like there's something that could be done with the .IO.e.
Like
next if none map ($file, RAKUDO-SRC ~ '/' ~ $file).IO.e
that would also pluck out the first one to exist.
Updated
is there gather/take or some other kind of to-me magic going on whereby p6 knows that we're done with the file at $lineno
No. I've concluded what I wrote was sloppily redundant -- I think my suggested code was mostly an improvement on what you had written but the use of a slurp for this was about me being insufficiently P6 fluent, not Perl 6 being lazy. To be specific:
- The slurp* routines are eager. The .slurp-rest method (which used to be called just .slurp) will suck up the (rest of the) entire file (or whatever) associated with the IO::Handle invocant.
- All the .lines methods are lazy. In my example code it's calling the .lines method on a string, the result of the slurp. It will stop when it reaches the $lineno line but that laziness comes after the non-lazy slurp has taken the time and RAM to slurp the whole file.
I think idiomatic lazy code would be something more like:
say "• ", $file.IO.lines[$lineno-1];
That drops the eager and unnecessary slurp and uses the lazy .lines method directly on the file handle returned by the .IO method called on $file. Aiui that will just read up to the $lineno line and then stop.
But I suspected that would leave the $file handle open. So instead I inserted a slurp, which I know opens and closes the handle, and then called Str's .lines method on its output. So my first suggestion was a sloppy slurpy version rather than a lightweight lazy one.
Since initially writing this comment I've started browsing the various lines method declarations in the source code of the existing IO::Handle module. This has led me to rewrite this comment and add these notes about what I see from the source code:
- I see what looks like bugs and a bit of a mess of variants of the .lines method. This may not matter because the newio branch is intended to replace this code. I plan to investigate this a bit further and comment in #perl6.
- There's a .lines method variant that takes an arg specifying the line number at which to quit reading. I think that's unnecessary for what you want but thought I'd mention it.
- There are also variants with a $close boolean arg which presumably avoid leaking open handles.
next if none map ($file, RAKUDO-SRC ~ '/' ~ $file).IO.e
I don't think that will compile. Also, nothing assigns to $file (for use with the subsequent slurp or whatever) if the RAKUDO-SRC prefix is needed. And if you included the missing assignment, the map would mean the assignment and test would happen even if it wasn't needed. Also, none
is a junction and junctions may do things in an unknown order (possibly concurrently) whereas your original logic was guaranteed to first test the file without the prefix and stick with that if it existed.
Maybe:
next unless first *.IO.e, $file, $file [R~]= RAKUDO-SRC ~ '/';
p6doc covers first
.
[R~] is the Reverse metaop applied to concat. It prepends instead of appends.
The A [op]= B
construction is the same as A = A op B
; the result of the operation is assigned back in to A (in this case A is $file
).
Getting even more crazy:
next unless first *.IO.e, $file, RAKUDO-SRC ~ '/' R[R~]= $file;
The outer R of the slightly insane R[R~]=
reverses the assignment. So, iff the unprepended $file doesn't exist, the RAKUDO-SRC ~ '/' on the left gets prepended to the variable ($file) on the right. :)
My word, @raiph, thanks for the write-up.
Yeah, knew that map
line wouldn't work (or even parse), just meant to illustrate where I was reaching.
And, dayum, never expected this to go where you've taken it. Hoped mind you, felt perhaps it could, but...wow.
I'm really starting to love perl6. A lot of that is b/c of folk who just jump in and explain the way you have.
On to the code...
Both list elts given to first
are evaluated, so $file
ends up always points to the rakudo source dir.
You can see it here
my $f = 'a';
say first * ~~ /a/, $f, $f [R~] 'b';
which yields ba
. That's b/c:
my $f = "a";
say ($f, $f [R~]= "b");
assigns to $f
and gets you ba ba
.
Maybe the 2nd elt can be thunked, somehow? So that the RAKUDO-SRC expression is only evaluated if first
gets to it?
As for $file.IO.lines()[$i]]
vs. looping through $file
by hand:
$ time
Well...speaking of time, I'm out of it. I'll update this comment with timings later today (I hope). But indexing into .IO.lines()
is way faster.
So, ~3x faster on my box:
$ time perl6 -e 'my $n = 15000 ; say "/usr/share/dict/words".IO.lines()[$n-1]; '
Tom
real 0m0.331s
user 0m0.306s
sys 0m0.024s
$ time perl6 -e 'my $n = 15000 ; my $fh = open "/usr/share/dict/words"; my $i = 0; for $fh.lines() { next unless ++$i == $n; .say; last } '
Tom
real 0m0.976s
user 0m0.947s
sys 0m0.028s
I'll be looking for your comments on #perl6 re: bugs in .lines()...but I might first try to find them on my own as an exercise.
Thanks again, this has been grand.
I'm really starting to love perl6. A lot of that is b/c of folk who just jump in and explain the way you have.
:)
Both list elts given to first are evaluated, so $file ends up always points to the rakudo source dir.
Oh! :)
say first * ~~ /a/, $f, $f [R~] 'b';
Note you could have just written:
say first /a/, $f, $f [R~]= 'b';
(Because Any method first
, which is called by sub first
, has a Regex variant.)
But yeah. I was thinking Rakudo wouldn't evaluate all the list elems because first
is lazy. But it does. So it's your turn. :) I'll return later to read what you've found out about this...
Interesting timing difference.
This stuff is all about list processing speed. And that will (hopefully) be profoundly altered by the recently begun GLR. Basically I recommend most folk don't worry about timing for this sort of stuff until August. If you want to focus on timing related to basic list operations because you find the news exciting, then pay close attention to GLR news.
Thanks again, this has been grand.
Yw. Please also feel free to post reddits and comments in /r/perl6. And of course, ask questions on #perl6. :)
The "next unless .. .IO.e" totally works.
Would the .slurp-rest read in the rest of the file? Or is there gather/take or some other kind of to-me magic going on whereby p6 knows that we're done with the file at $lineno and after?
Well, I suppose I'll find out.