egg|zzz|egg changed the topic of #principia to: READ THE FAQ: http://goo.gl/gMZF9H; The current version is Chasles. We currently target 1.2.2, 1.3.0, and 1.3.1. <scott_manley> anyone that doubts the wisdom of retrograde bop needs to get the hell out | https://xkcd.com/323/
<xShadowx>
blowfish does anyways
e_14159 has quit [Ping timeout: 198 seconds]
e_14159 has joined #principia
stratochief_ is now known as stratochief
icefire has quit [Read error: Connection reset by peer]
egg is now known as egg|zzz|egg
Wetmelon has quit [Ping timeout: 183 seconds]
<awang>
egg|zzz|egg: How many test failures should I be seeing?
<awang>
If any?
Wetmelon has joined #principia
Virindi has joined #principia
Virindi has left #principia [#principia]
<egg|cell|egg>
None?
<egg|cell|egg>
What are you seeing
<awang>
Uh
<awang>
Hold on
<awang>
Something like ~20 test failures, IIRC
<awang>
egg|zzz|egg: Does the test thing automatically generate a log file? Or do I need to pipe the output?
ferram4__ has quit [Read error: Connection reset by peer]
ferram4__ has joined #principia
<awang>
egg|zzz|egg: 38 failed tests
<awang>
I'll get a Dropbox'd file linked
<awang>
too much to paste
<awang>
Also, sort of curious
<awang>
How much do you think -ffast-math would mess with things?
<egg|zzz|egg>
awang: um how did you get a test failure in doubleprecision
<egg|zzz|egg>
that's bad
<egg|zzz|egg>
awang: did you change the options from the usual configuration?
<egg|zzz|egg>
awang: you're getting plenty of failures in numerics code that didn't change; that's really not good.
<egg|zzz|egg>
lamont: you had no such failures right?
<awang>
egg|zzz|egg: The dropbox'd log was from eggrobin/use-the-downsampling
<awang>
No changes made
<awang>
ffast-math was more of a curiosity. I'm not that dumb :P
<egg|zzz|egg>
awang: well it's really weird
<egg|zzz|egg>
awang: does it also have those errors at Chasles
<awang>
Pretty sure those failures have been around a while
<awang>
Just haven't gotten around to asking you about them until now
<egg|zzz|egg>
awang: also are there any fslow-math that you could enable so it goes into "do what it says damn it" mode
<awang>
Um
<awang>
I think that's what it does by default with -O3?
NolanSyKinsley has quit [Remote host closed the connection]
<egg|zzz|egg>
so the DoublePrecision consistency checks should pass if that's the case, no matter what libm does (which is the main source of discrepancy)
<egg|zzz|egg>
so something is very fishy if they don't
<egg|zzz|egg>
oh wait it's a test that compares debugstrings
<egg|zzz|egg>
yeah those might not be platform-agnostic unfortunately
<egg|zzz|egg>
we should write our own double-to-string >_<
<egg|zzz|egg>
ok then sgtm
<awang>
Hmmm
<awang>
It doesn't compare the double values directly?
<egg|zzz|egg>
I don't remember why, but not here no
<egg|zzz|egg>
maybe it should tbh >_>
<egg|zzz|egg>
but anyway, ignore the test failures for now
<awang>
How do you convert to DebugStrings?
<awang>
Should I open an issue?
<egg|zzz|egg>
(most of them will be libm-dependent and that's platform-dependent and I trust Atlas to know how to into libm)
<egg|zzz|egg>
awang: maybe
<awang>
Atlas?
<awang>
And I thought that the standard has precision requirements on math functions?
<awang>
Oops, yeah, standard doesn't require precision stuff
<awang>
IEEE does
<egg|zzz|egg>
recommends, not require
<egg|zzz|egg>
requires for sqrt, and that you generally have
<egg|zzz|egg>
transcendental functions you almost never have
<egg|zzz|egg>
(except with libmcr as fiora pointed out)
<awang>
libmcr?
<egg|zzz|egg>
!g "libmcr"
<Qboid>
egg|zzz|egg: https://github.com/simonbyrne/libmcr [GitHub - simonbyrne/libmcr: Sun libmcr (correctly rounded libm)] (69 results found, took 0.35s)
<awang>
Oh huh
<egg|zzz|egg>
(basically a libm that does arbitrary-precision in edge cases to be correctly-rounded always
<egg|zzz|egg>
and now you understand why in general you don't use that :-p
<awang>
Don't use what?
<egg|zzz|egg>
libmcr
<egg|zzz|egg>
slow in edge cases, doesn't gain much
<awang>
Ah
<awang>
Sorry, still new to the entire numerical everything
<egg|zzz|egg>
ideally you want a ~faithfully rounded libm, with bounds on how often it goes to the wrong side
<egg|zzz|egg>
awang: ask around in kspacademia, that's what it's here for :-p
<egg|zzz|egg>
that and catpics
<awang>
And unicode :P
<egg|zzz|egg>
but then catpics can have numerical tips so :-p
<awang>
Ah, so that's why the glibc page on max errors is so long
<GH>
[Principia] aw1621107 opened issue #1624: Test failures on macOS https://git.io/vFwQ0
ferram4__ has quit [Ping timeout: 186 seconds]
ferram4__ has joined #principia
egg|phone|egg has joined #principia
egg|cell|egg has quit [Read error: Connection reset by peer]
egg|cell|egg has joined #principia
egg|mobile|egg has joined #principia
egg|cell|egg has quit [Read error: Connection reset by peer]
egg|phone|egg has quit [Ping timeout: 383 seconds]
egg|mobile|egg has quit [Ping timeout: 198 seconds]
egg|phone|egg has joined #principia
egg|cell|egg has joined #principia
egg|phone|egg has quit [Read error: Connection reset by peer]
<GH>
[Principia] eggrobin commented on issue #1624: With the exception of `DoublePrecisionTest.Consistencies`, all those tests have libm-dependent tolerances, so the failures are expected (assuming that the tolerances do not change massively).... https://git.io/vFwxW
<GH>
[Principia] pleroy closed pull request #1622: Use the downsampling (master...use-the-downsampling) https://git.io/vFV1A
<GH>
[Principia] pleroy closed issue #228: Keep the whole history https://git.io/0Ep-1w
<egg|zzz|egg>
awang: just to check obvious things, you *are* compiling for x64 with sse2, not x87 floating-point, right?
<GH>
[Principia] pleroy closed issue #1618: crash on minmus https://git.io/vFCbY
<GH>
Principia/master 6bf8a5e Robin Leroy: #1618 is the new #1447
egg|phone|egg has joined #principia
egg|cell|egg has quit [Ping timeout: 198 seconds]
egg|cell|egg has joined #principia
egg|phone|egg has quit [Ping timeout: 200 seconds]
egg|phone|egg has joined #principia
egg|cell|egg has quit [Ping timeout: 183 seconds]
egg|cell|egg has joined #principia
egg|phone|egg has quit [Ping timeout: 186 seconds]
ferram4__ has quit [Ping timeout: 198 seconds]
ferram4 has joined #principia
egg|phone|egg has joined #principia
egg|cell|egg has quit [Ping timeout: 198 seconds]
egg|cell|egg has joined #principia
egg|phone|egg has quit [Read error: Connection reset by peer]
egg|phone|egg has joined #principia
egg|cell|egg has quit [Read error: Connection reset by peer]
egg|phone|egg has quit [Read error: Connection reset by peer]
egg|phone|egg has joined #principia
<lamont>
awang: egg|zz|egg: yeah there have been failures for quite some time. they should all be on long integration tests and should be small numbers of ULPs tho
<lamont>
i asked you about it a long time ago, and we came to the conclusion they were ignorable
<egg|zzz|egg>
lamont: yeah, someday I'd like to investigate the double-double one, because it looks nonignorable; the others are different libms having different results, WAI
<egg|zzz|egg>
(we could ifdef tolerances by libm if we were nuts)
<egg|zzz|egg>
(or ship our own libm if we were nuttier)
<lamont>
lol
<egg|zzz|egg>
lamont: any opinions on deprecating 1.3.0?
<lamont>
1.3.0?
<egg|zzz|egg>
lamont: or at the nuttiest, using libmcr! :D who cares about performance when you can have correctly-rounded everything rather than faithfully-rounded :D
<lamont>
lol, you need to care about perf at least a little bit
<egg|zzz|egg>
hence the "nuttiest" bit :D
<egg|zzz|egg>
lamont: we'll be shipping the upcoming release for 1.2.2, 1.3.0, and 1.3.1; I'm tempted to make it the last one that supports 1.3.0, see the RFC in #RO
<lamont>
i would drop support for any old patch releases, only track the latest minor of the last two revs
<egg|zzz|egg>
lamont: yeah
<egg|zzz|egg>
lamont: we had the last release right after 1.3.1, so it made sense to still support 1.3.0
<lamont>
people on RO/RSS are stuck on minor-1, but people on stock usually upgrade fairly aggressively.
<lamont>
yeah i mean if its like the day after 1.3.1 drops, then you could do the people on 1.3.0 a solid since most likely most other mods wouldn’t have updated, but i’d drop it the release after
<egg|zzz|egg>
lamont: yeah and RO will likely skip 1.3.0
<lamont>
yes
<lamont>
it’ll probably go directly to 1.3.3 the week after 1.4.0 drops...
<egg|zzz|egg>
lamont: well I don't want to drop 1.3.0 support for this release because the update reminder is there for them too without a special mention, so I need to plan deprecations a moon in advance :D
<egg|zzz|egg>
lamont: wait there's an 1.4 in the works?
<egg|zzz|egg>
or is it just a joke on the glacial pace :-p
<lamont>
that was more of a joke on the glacial pace, but 1.4.0 is being worked on
<lamont>
it’ll have a unity bump, it got announced a few weeks ago, i’d guess 4-6 months off or whatever
<egg|zzz|egg>
ah okay, we might be off the C releases by then even :D
<lamont>
hah
<egg|zzz|egg>
lamont: does FAR work on 1.3.1? the thread title still says 1.3.0
<lamont>
i still don’t think i’ve ever fired up 1.3.x
<egg|zzz|egg>
:D
NolanSyKinsley has joined #principia
ferram4 has quit [Read error: Connection reset by peer]
ferram4 has joined #principia
Jesin has joined #principia
<GH>
[Principia] eggrobin opened pull request #1625: Journal for 陈景润 (master...journal-for-陈景润) https://git.io/vFr3R
ferram4_ has joined #principia
ferram4 has quit [Read error: Connection reset by peer]
Hypergolic_Skunk has quit [Quit: Connection closed for inactivity]