dalek arVM: a7b842c | (Jimmy Zhuo)++ | src/6model/reprs/P6bigint.c:
small cleanup
 03:15
04:13 btyler joined 04:15 harrow joined 04:46 harrow joined 05:34 harrow joined 07:15 TimToady joined 07:19 mj41 joined 07:57 FROGGS joined 08:04 kjs_ joined 08:09 woolfy joined 08:10 woolfy left 08:16 rurban_ joined 08:40 dalek joined 08:47 zakharyas joined 09:01 brrt joined 09:02 zakharyas1 joined 09:09 kjs_ joined
dalek arVM/native-ref: 5aed474 | jnthn++ | / (6 files):
Stub in native reference related ops.
 09:40
09:54 btyler joined 10:16 avar joined
dalek arVM: c0bae2e | lizmat++ | / (6 files):
Add readlink stub
 10:38
jnthn lol, the last two will be a merge conflict between them... :) 10:39
nwc10 I was just thinking something equivalent when reading Liz's commit - it's a shame that there are massive diffs as a side effect of moving auto-generated things around 10:40
I'd not thought of conflicts
lizmat jnthn: sorry, but it should be easily fixed, no? Or do you want me to revert that last one until you merged ? 10:52
jnthn lizmat: I teach advanced git courses, merge conflicts are the least of my problems. :) 10:53
lizmat ok, *phew*, don't want to put more on your plate
jnthn Was just amused at the timing; changes to src/core/oplist are relatively rare, then two come along at once :) 10:54
brrt jnthn - can you come by and teach $company at my $day-job even starting git techniques 10:58
they're all like 'whats a remote' over here 10:59
arnsholt My boss is still very suspicious of git. Doesn't want to fall prey to hype-chasing =) 11:00
brrt lol
jnthn Hype-chasing? It's been almost 10 years... ;)
arnsholt Yeah, I know
brrt lunch & 11:01
arnsholt So he's still on SVN
Thankfully most of his repos I can just use git-svn on =D
jnthn API design is harder than merge conflicts. Hmmmm... 11:02
nwc10 arnsholt: when it comes to doing the right thing to svn, I'm finding that techbase.kde.org/Projects/MoveToGi...ingSvn2Git is pretty damn good 11:13
(used git-svn for the previous humane termination of svn and it was not fab)
and oh my, svn...
it was 90 minutes to import a svnadmin dump of our 40,000 commit repo into a clean repo on my work desktop (Which has an SSD) 11:15
and I think about 15 for svn2git to import the entire bloody thing into git
one of these is the native tool
one of these is not
subversion.apache.org/roadmap.html - Release 1.9.0 - Q2 2014 (tentative)
not as late as Xanudu, Duke Nukem Forever or Perl 6 11:16
11:16 kjs_ joined
arnsholt nwc10: And svn2git lets me interact with the SVN repo too? 11:16
nwc10 I find it impossible to search for succesfully, but I know I once read a blog entry about how new svn committers get voted in once a year
and they realised they were in a bad place when that year the list to vote on was empty
arnsholt: "when it comes to doing the right thing to svn" => "when you are allowed to kill it" 11:17
no, it doesn't
but it looks to be better for a final conversion
arnsholt Yeah, that's what I thought
nwc10 svn was awseome. For a bit.
Then better things arrived
arnsholt Getting rid of it is currently not an option
The one I'm working with ATM, I might be able to get rid of 11:18
nwc10 but oh my, having had cause to look at the code of both recently the architecture of svn is "no obvious bugs"
git is "obviously no bugs"
and then you wonder why git is so fast
because it's "the simplest thing that could possibly work", and almost no levels of abstraction
(sorry if this sounds like a mix between a rant and a marketing spiel)
arnsholt The other one, ye olde repo of doom, with a pile of externals and 15 or 20 years of history, not so much
nwc10 but I'm finding so many little frustrations with the svn tooling 11:19
svnlook vs svn
arnsholt 25G logon/
nwc10 repos vs checkouts
arnsholt On a 110G disk >.<
Yeah, it's death by a thousand cuts
nwc10 no easy way to get some bits of data out 11:20
arnsholt No one thing that really kills it, but the overall experience is just terrible
nwc10 although I realise now, it feels like svn is a library you code against in C
git is a set of tools you glue together in shell
(or maybe someone wrote something better than shell. any suggestions?)
arnsholt Heh
nwc10 and the latter is far better for both Manipulexity and Whipuptitude 11:21
arnsholt The UI of git is pretty crazy, but the feature-set is awesome
nwc10 arnsholt: once you've proven a conversion works on the smaller repository
can you chip away at the big one subtree by subtree? 11:22
arnsholt Might be possible
Being the repo of doom, it really does want to be several smaller repos
(And the bundled Java 1.4 or something like that *really* has to go)
nwc10 but Java is Java - surely the code can run on a newer JVM? :-) 11:23
arnsholt Of course it does 11:24
But they had some kind of distribution pain at some point I think, and Java not being as widely available 15 years ago
So they just bundled the whole damn thing
And once it's in, it's not going out (obv)
Oh, wait. It's Java 6, not 4 11:25
nwc10 yay.
arnsholt But still messed up my NQP/JVM hacking, since the bundled "set things up" bashrc file *prepends* to $PATH 11:26
Did I mention that I don't like all of my boss's techincal decisions? =)
nwc10 so that would be the analagous problem to "I'd like to write code that uses features in v5.12, but /usr/bin/perl is only 5.8.8"?
arnsholt Sort of
nwc10 arnsholt: does your current boss like all of his previous self's choices? :-)
arnsholt He's the keeper of the repo, so it's perfectly tailored to his workflows 11:27
So he obviously doesn't see any problems with this approach =p
lizmat argh, I interpreted the sig wrong in the oplist :-( 11:34
dalek arVM: c50bd77 | lizmat++ | / (3 files):
Fix prototype on readlink stub
 11:37
11:55 Ven joined 11:57 xiaomiao joined 11:59 ggoebel111111113 joined 12:02 arnsholt joined 12:04 lizmat_ joined
FROGGS how does readlink work? it just reads a reads and returns nothing? 12:08
lizmat: ^^ 12:09
lizmat well, no, it should return a string
or fail, presumably
FROGGS +readlink w(str) r(str) then
lizmat so the w is for the input, and the r for the output? 12:10
FROGGS w is writing == returning
12:10 dalek joined
FROGGS like in: $w = readlink($r) 12:10
that's why the thing you return comes first
12:10 moritz joined
lizmat ok 12:11
lizmat is learning
(I hope)
FROGGS *g*
12:12 [Coke]_ joined 12:13 sivoais joined 12:14 rurban__ joined
dalek arVM: c27b31d | lizmat++ | / (3 files):
Hopefully got the readlink sig ok now
 12:14
12:15 woolfy joined
FROGGS looks good to me 12:17
12:21 btyler joined, japhb joined
lizmat so I need to return an MVMString from what appears to be a void *: 12:46
void* uv_fs_t.ptr
Stores the result of uv_fs_readlink() and serves as an alias to statbuf.
I'm not sure how to do that
FROGGS jnthn suggestions?
FROGGS gimme a sec 12:47
lizmat (MVMString *)req.ptr; # seems too succinct :-) 12:48
FROGGS no, that's quite wrong :o)
to check that uv_fs_t.ptr is a C string, I'd try to printf it first... like in: fprintf(stderr, "req.ptr = '%s'\n", (char *)req.ptr); 12:50
okay, according to linux.die.net/man/2/readlink you get a C string, and I guess that libuv null-terminates the string from readlink(2) for us 12:52
lizmat so how do I convert a C string to a MVMString ? 12:53
FROGGS * a result of the specified type. The type must have the MVMString REPR. */ 12:54
MVMString * MVM_string_ascii_decode_nt(MVMThreadContext *tc, MVMObject *result_type, const char *ascii) {
err
* Decodes a NULL-terminated ASCII string into an NFG string, creating
* a result of the specified type. The type must have the MVMString REPR. */
MVMString * MVM_string_ascii_decode_nt(MVMThreadContext *tc, MVMObject *result_type, const char *ascii) {
lizmat Hmm... are we sure it's ASCII we get?
I would assume UTF8, no? 12:55
jnthn I'd go with UTF8 for now 12:56
lizmat MVM_string_utf8_encode_C_string then?
jnthn Mostly 'cus I think that's what happens elsewhere
No, that's the wrong directly
FROGGS MVM_string_utf8_decode(tc, type_object, Cbuf, byte_length)
jnthn encode = MVMString => bytes (e.g. C string) 12:57
That's the one
Just search for existing usages of it.
FROGGS lizmat: also note that creating an MVMString means that you are creating an object that our GC cares about
creating (and therefore allocating) such a thing might directly trigger a GC run 12:58
jnthn Thankfully, it's probably also being done as the very last thing
FROGGS ... which will have nice side effects
jnthn So it's easy this time
lizmat yeah, the last thing
jnthn But yes, you do have to be very aware of the GC when working on Moar
FROGGS would just be nice to review a gist before committing 12:59
jnthn I really should write a MoarVM hackers guide at some point
lizmat gist.github.com/lizmat/c68fda121ea7078e52cc
only the "type_object" I'm not sure what to put ? 13:00
and is the strlen ok?
jnthn Surely you're missing a return ?
FROGGS that's not perl :D
jnthn lizmat: tc->instance->VMString for type_object
FROGGS jnthn: I don't remember... do we free things we declare const? 13:01
jnthn And yes, srlen
FROGGS srlen?
jnthn strlen :)
lizmat updated the gist
jnthn
.oO( strlen steak )
.oO( I really should go shopping for lunch... )
 13:02
lizmat visitors, afk&
jnthn lizmat: Missing ; on end of last line, otherwise looks sane
FROGGS - if (uv_fs_readlink(tc->loop, &req, path_s, 0, NULL)) { 13:03
+ if (uv_fs_readlink(tc->loop, &req, path_s, NULL)) {
I think
lizmat yeah, found that just now, FROGGS++ 13:04
and jnthn++
FROGGS ohh, you have to care about readlink's return value
jnthn Ain't that what the if is doing?
FROGGS On success, readlink() returns the number of bytes placed in buf. On error, -1 is returned and errno is set to indicate the error.
jnthn Oh... 13:05
FROGGS so, you have to record that and pass it off instead of strlen()
lizmat ah.... ok
jnthn So you can avoid the strlen
Well, that refactor is also good in that you can get rid of the duplicate MVM_free(c_str); too
By putting the result into a variable, do the free, then look at it.
shopping & 13:06
FROGGS I guess we want to put the text of the possible error codes in an exception?
linux.die.net/man/2/readlink
but, that can be put in after it works
lizmat gist updated again 13:10
FROGGS looks good 13:11
I guess an empty string will never happen? 13:12
as the result of readlink I mean
lizmat then I guess we have a failure 13:13
FROGGS probably 13:14
dalek arVM: e98ea25 | lizmat++ | src/ (3 files):
Hopefully final tweaks to get readlink working

  jnthn++ FROGGS++ for assistance		 13:24
FROGGS I always enjoy implementing ops in MoarVM 13:27
lizmat++
lizmat $ ./nqp -e 'say(nqp::readlink("/tmp"))' 13:28
Failed to readlink file: Unknown system error
:-(
lunch&
arnsholt lizmat: That should probably fail with some kind of error though, unless your /tmp is a symlink 13:31
FROGGS it should say "The named file is not a symbolic link. " 13:32
I guess we cannot use uv_strerror here
nwc10 if you want to test error messages, force a locale 13:33
else you get confused users and bug reports
FROGGS ohh, and also, it needs to be "if (pathlen >= 0) {" 13:34
13:55 zakharyas joined
FROGGS $ nqp-m -e 'say(nqp::readlink("/tmp"))' 14:11
Memory allocation failed; could not allocate 18446744073709551528 bytes
jnthn Whyever not? It's only 18 exabytes... 14:14
FROGGS that's about a gazillion times Data's brane 14:15
In "The Measure of a Man", a Starfleet judge rules that Data is not Starfleet property. The episode establishes that Data has a storage capacity of 800 quadrillion bits, (88.81784197 PiB) and a total linear computational speed of 60 trillion operations per second. 14:16
nwc10 or -87
Peter_R FROGGS, how do those op/s measure up to modern machines? 14:31
FROGGS Peter_R: looks like an Intel Core i7 5960X does 336_000 operations per second 14:35
so, Data is quite good
ohh, wait 14:36
it is 336_000_000_000 instructions per second
jnthn Well, if you're doing a cycle = instruction approximation. :) 14:37
In reality, an instruction might take hundreds of cycles, or one cycle might process several instructions
nwc10 and the best video I've seen on that (so far) is www.infoq.com/presentations/click-c...n-hardware 14:38
FROGGS the number I've shown is calculated using the Dhrystone benchmark
we should ask Brent Spiner to do that too :o)
lizmat arnsholt: on OS X, /tmp is a symlink to /private/tmp 14:41
FROGGS lizmat: I have a patch... 14:49
lizmat: can I commit or are you working on it?
nwc10 and a test for readlink on a known dir, known file, and nowknown missingnes? 14:50
lizmat FROGGS: please patch :-)
nwc10 (am I being unhelpful?)
FROGGS nwc10: slightly
lizmat nwc10: well volunteered :-)
FROGGS because windows and filesystems and uff
nwc10 IIRC for Perl 5, readlink dies on systems that don't have the syscall 14:51
part of me is tempted to think that it's arguably also more logical for it just to return failure
dalek arVM: a910556 | FROGGS++ | src/io/fileops.c:
fix readlink, the return value is just an error indicator
nwc10 (ie nothing is a symlink on an OS that doesn't support them, rather than it being an excepton to ask)
but my brain can argue it both ways
lizmat: there's another bunch of stuff in my stack that hopefully gets unblocked before that 14:52
FROGGS nwc10: I suspect we will return a Failure from P6 land, aye
lizmat FROGGS: I guess my original approach was best :-) 14:59
FROGGS :o) 15:01
the problem is just that pathlen was 0 on success
so it is not the ssize_t the readlink(2) returns 15:02
libuv's documentation quite sucks here
lizmat $ ./nqp -e 'say(nqp::readlink("/tmp"))'
private/tmp
YEAH!
ah, relative to the original path 15:03
ok, I can live with that :-)
FROGGS me too :o)
MoarVM$ nqp-m -e 'say(nqp::readlink("../p6-Text-CSV/Text/CSV.pm"))' 15:04
../test-t.pl
MoarVM$ nqp-m -e 'say(nqp::readlink("VERSION"))' 15:05
Failed to readlink file: invalid argument
MoarVM$ nqp-m -e 'say(nqp::readlink("foobarbaz"))'
Failed to readlink file: no such file or directory
I think that's okay
15:07 kjs_ joined
dalek arVM: e0c9bb0 | (Jimmy Zhuo)++ | src/6model/reprs/P6bigint.c:
fix comment style, FROGGS++
 15:07
15:30 Ven joined
dalek arVM/native-ref: 13a893c | jnthn++ | src/core/callsite. (2 files):
Add some extra callsites.

Will be needed by native fetch/store of the code pair container spec.		 15:53
arVM/native-ref: db78722 | jnthn++ | src/6model/containers. (2 files):
Extend container API for native handling.

This also updates the code_pair container spec to match the extended API.
arVM: 4e79d84 | (Jimmy Zhuo)++ | src/io/fileops.c:
free req.ptr in MVM_file_readlink
 16:04
jnthn JimmyZ++ # nice catch 16:06
dalek arVM/native-ref: 6b4f3fc | jnthn++ | src/ (3 files):
Implement decont_[ins].
 16:09
JimmyZ jnthn: raw.githubusercontent.com/rakudo/r...ontainer.c 16:11
jnthn: I didn't see any get_mu codes, Did I miss something?
and get_nil() ...
jnthn JimmyZ: Urgh. But I think they're actually implemented in perl6_ops.c 16:12
Barely legal XXX C!
It's exactly what we do in .h files, though.
Declare "there's a function like this somewhere and you'll get it by link time" :) 16:13
JimmyZ yeah, the name make me thinks it's static :P 16:14
jnthn Uh, yeah...we probalby should not have a non-static thing with an unprefux name...
dalek arVM/native-ref: 94dec40 | jnthn++ | src/ (3 files):
Implement assign_[ins].
 16:18
16:52 synopsebot joined 17:01 sivoais joined
TimToady you guys do know that filenames are not guaranteed to be utf8? 17:10
TimToady overnighting in Medford, OR
s/ing/ed/
jnthn TimToady: Yes, but it's an OK first approximation for now, and it may be one of those places where NFG might help us (in so far as we can have a scheme where anything that ain't UTF-8 is handled as a synthetic when we make the thing a Str, so it round-trips) 17:13
And I'd rather the code consistently gets it not right so it's easy to grep all of those out.
TimToady I guess that's as good a story as any
the alternative is to return either utf8 or blob8 and let the user deal with it, I suppose 17:15
jnthn I'm pretty sure the "make everyone use Buf for filenames" altnerative isn't going to fly with the majority of programmers out there...
Since "filenames are strings" is pretty deeply embedded in developer culture everywhere... 17:16
TimToady well, synthetics composed of illegals is also a discontinuity, just elsewherely placed 17:19
but maybe the concept is useful generally
17:26 FROGGS[mobile] joined
arnsholt jnthn: native-ref is towards fixing the $native++ problem, right? 17:26
17:26 FROGGS[mobile]2 joined
jnthn Yes, and is rw accessors for native attributes, and making native arrays work, and... :) 17:27
FROGGS 'is rw' ftw!
bbiab
arnsholt Indeed. Spiffy stuff going on these days!
jnthn
.oO( Why solve one problem when you can solve 3 at once... )
arnsholt Quite
Both this and the parametric stuff is good for NativeCall too
jnthn Anyway, figured out how to do most of the pieces needed. 17:28
So at the moment working through implementing them.
Though, I notice I'm rather hungry so maybe I should pause for dinner :)
TimToady do native refs know the limit of what they're refing? or are they always singular?
jnthn TimToady: Always singular at the moment, and naively they go via the same API as something doing a direct update of the thing would. 17:30
TimToady: However, I'm designing it so spesh will be able to "see through" that safety
17:30 ingy1 joined
jnthn TimToady: Not to mention get rid of the taking/using of native references completely when inlining renders it uninteresting. 17:31
TimToady: For cases like $native-int++ I intend the static optimizer nail those.
The dynamic case is more about when you inline an accessor method for a native attribute. 17:32
TimToady what about visiting a slice of an array? 17:36
thinking about Go's slice-refs here... 17:37
jnthn Example Perl 6 code that's use what you're thinking of? 17:38
(So I can make sure we're talking about the same thing... )
*that'd 17:39
TimToady can you ++ a native ref (not the thing referred to, but the pointer)?
TimToady doesn't know what a native pointer looks like in Perl 6 yet... 17:40
in Go I think you can ++ a slice and it'll do bounds checking for you 17:41
jnthn Ah
No, don't have anything for that yet
TimToady it seems like an efficiency thing to have bounded pointers
jnthn But don't think the approach I'm taking will make it impossible to add that later.
For now I'd like native arrays to work at all ;) 17:42
Then we can get clever. 17:43
TimToady okay, just bear it in mind so you don't make it impossible later, thanks
timotimo jnthn: i'd like to hear how you imagined the "please compile an accessor method for me" thing to be implemented; i.e. which part does what where 17:54
17:59 lue joined
jnthn timotimo: OK...but dinner is ready, so after that :) 18:03
bbiab
timotimo OK 18:04
i think i want to get some food my own self
18:10 tgt joined 19:22 kjs_ joined
[Coke] ugh, no food. :P 19:32
jnthn Can't eat? :( 19:34
That's the wurst thing...
dalek arVM/native-ref: 3a55063 | jnthn++ | / (8 files):
Start getting NativeRef REPR fleshed out.

We put the kind of referenced thing and the primitive type as REPR data. This means there'll be a minor explosion of types for native references, but they align well with the way we'll want to do type specialization, so it's worth it. They'll also be invisible to most users (except those who go digging).		 19:39
19:50 FROGGS joined
[Coke] retroactively UGHS at jnthn++'s pun 20:39
20:51 kjs_ joined 21:00 rurban__ joined
dalek arVM/native-ref: b268435 | jnthn++ | src/core/hll.c:
Needn't NULL stuff with we calloc.
 21:00
arVM/native-ref: b989127 | jnthn++ | src/core/hll. (2 files):
Add HLLConfig entries for native ref types.
21:06 ingy joined
dalek arVM/native-ref: f951a73 | jnthn++ | src/ (3 files):
Validate native references when adding to HLL.

This means we can skip any checks later on and Just Use Them.		 21:25
arVM/native-ref: f7b7985 | jnthn++ | src/6model/reprs/NativeRef.c:
Serialization bits for NativeRef REPR data.
arVM/native-ref: c76e414 | jnthn++ | src/6model/containers.c:
Stub native_ref container spec.
 21:40
22:04 FROGGS_ joined 22:27 colomon joined 22:29 colomon joined 22:37 woolfy left 22:45 nebuchadnezzar joined 23:15 colomon joined 23:34 colomon joined

Powered by Cro and the Raku® Programming Language.

Please report any issues / comments / feature requests as an issue on App::Raku::Log.

Thank you!