all: add support for FreeBSD 11 #7849

gopherbot · 2014-04-23T18:09:08Z

by smith.winston.101:

Building Go 1.3-beta1 on FreeBSD-CURRENT-r264670/armv6 (Beaglebone Black) fails the
following tests (see below for details):
    1) math
    2) net
    3) strconv

Go 1.3beta1 was built as follows:

CC=clang ./all.bash


`go version` shows:

go version devel +f8b50ad4cac4 Mon Apr 21 17:00:27 2014 -0700 + freebsd/arm


`clang --version` shows:

FreeBSD clang version 3.4 (tags/RELEASE_34/final 197956) 20140216
Target: armv6--freebsd11.0-gnueabi
Thread model: posix


NOTES:
    1) Go 1.3-beta1 compiled and ran tests OK on FreeBSD-RELEASE-10 using Clang 3.3 (tags/RELEASE_33/final 183502)
    2) Go 1.3-beta1 compiled and ran tests OK on the same BeagleBone Black hardware running Linux (Debian Wheezy, GCC 4.6.3).



--- FAIL: TestFrexp (0.00 seconds)
        all_test.go:2048: Frexp(0) = 0, 0, want 0.9999999999999998, -1022
        all_test.go:2048: Frexp(0) = 0, 0, want 0.5, -1073
        all_test.go:2048: Frexp(0) = 0, 0, want -0.9999999999999998, -1022
        all_test.go:2048: Frexp(0) = 0, 0, want -0.5, -1073
--- FAIL: TestIlogb (0.00 seconds)
        all_test.go:2108: Ilogb(0) = -2147483648, want -1023
        all_test.go:2108: Ilogb(0) = -2147483648, want -1074
        all_test.go:2108: Ilogb(0) = -2147483648, want -1023
        all_test.go:2108: Ilogb(0) = -2147483648, want -1074
--- FAIL: TestLdexp (0.00 seconds)
        all_test.go:2169: Ldexp(0.9999999999999998, -1022) = 2.225073858507201e-308, want 0
        all_test.go:2169: Ldexp(0.5, -1073) = 5e-324, want 0
        all_test.go:2169: Ldexp(-0.9999999999999998, -1022) = -2.225073858507201e-308, want 0
        all_test.go:2169: Ldexp(-0.5, -1073) = -5e-324, want 0
        all_test.go:2169: Ldexp(2.2250738585072014e-308, -52) = 5e-324, want 0
        all_test.go:2169: Ldexp(0, 1074) = 0, want 1
        all_test.go:2169: Ldexp(1.7976931348623157e+308, -2097) = 1e-323, want 0
--- FAIL: TestLogb (0.00 seconds)
        all_test.go:2229: Logb(0) = -Inf, want -1023
        all_test.go:2229: Logb(0) = -Inf, want -1074
        all_test.go:2229: Logb(0) = -Inf, want -1023
        all_test.go:2229: Logb(0) = -Inf, want -1074
--- FAIL: TestNextafter (0.00 seconds)
        all_test.go:2314: Nextafter(0, -1) = -5e-324 want 0
        all_test.go:2314: Nextafter(-0, 1) = 5e-324 want 0
        all_test.go:2314: Nextafter(-0, -1) = -5e-324 want 0
--- FAIL: TestFloatMinMax (0.00 seconds)
        all_test.go:2577: Sprint(SmallestNonzeroFloat64) = 0, want 5e-324
        all_test.go:2577: Sprint(SmallestNonzeroFloat32) = 0, want 1e-45
FAIL
FAIL    math    1.036s



--- FAIL: TestInterfaces (0.00 seconds)
        interface_test.go:59: table: len/cap = 2/2
        interface_test.go:71: InterfaceByName("") failed: invalid network interface name
FAIL
FAIL    net     5.944s



--- FAIL: TestFtoa (0.01 seconds)
        ftoa_test.go:149: testN=64 0 g -1 want 2.225073858507201e-308 got 0
        ftoa_test.go:149: AppendFloat testN=64 0 g -1 want abc2.225073858507201e-308 got abc0
        ftoa_test.go:147: testN=32 0 g -1 want 2.225073858507201e-308 got 0
        ftoa_test.go:151: AppendFloat testN=32 0 g -1 want abc2.225073858507201e-308 got abc0
FAIL
FAIL    strconv 6.168s

ianlancetaylor · 2014-04-23T21:14:51Z

Comment 1:

Labels changed: added repo-main, release-go1.3.

minux · 2014-04-23T21:39:01Z

Comment 2:

i know the reason for math and strconv:
freebsd-current uses runfast mode of vfp by default, which flushes denormal fp numbers
to zero. Our toolchain needs to be able to reprent denormal numbers faithfully for some
of the math/strconv tests.
I'm not sure about the net test failure though.

minux · 2014-04-23T21:43:57Z

Comment 3:

could you please give this CL:
https://golang.org/cl/89610043
a try to see if it could fix the math and strconv test failures?

Owner changed to @minux.

Status changed to WaitingForReply.

gopherbot · 2014-04-24T02:17:26Z

Comment 4 by smith.winston.101:

I rebuilt and retested on the BBB using FreeBSD-10-STABLE (r264836) -- Go 1.3beta1
builds and tests OK.  I'll retry 11-CURRENT with the patches in #3.

minux · 2014-04-24T02:22:42Z

Comment 5:

great! that means freebsd-11 does make some incompatible changes compared to 10.
thank you for the confirmation.

gopherbot · 2014-04-25T01:56:39Z

Comment 6 by smith.winston.101:

@minux, your CL RESOLVED the math and strconv issues on FreeBSD-CURRENT.
There's still an issue with the net package failing interface_test; it looks like the
results from syscall.ParseRoutingSockaddr() returns a zero length list -- I'm digging
into that, hopefully, I'll figure out what's going on (I can see the if names when I
hex.Dump() out the InterfaceMessage.Data blocks).

gopherbot · 2014-04-25T02:00:34Z

Comment 7 by smith.winston.101:

FYI: In my original report I wasn't clear that in NOTE #1 I was actually referring to
testing Go1.3beta1 on an x64 FreeBSD-10-RELEASE system.  Comment #4 was actually when I
tested FreeBSD-10-STABLE on the armv6 (BeagleBone Black).

minux · 2014-04-25T02:18:58Z

Comment 8:

Thank you for testing the CL!
I will remove irrelevant part of CL and get the essential fix included in 1.3 beta 2.
I currently don't know why TestInterface fails on freebsd-current/arm, so if you find
anything, please comment, thank you very much!
(mikio mentioned that it could be due to "a side effect of multicast and sockaddr_dl
cleanups on 11-current." in a recent post "[golang-dev] add support for FreeBSD 10")

mikioh · 2014-04-25T22:36:39Z

Comment 9:

Thanks for the testing Go 1.3-beta1 on FreeBSD 11-CURRENT.  We now understand that
freebsd11 has more surprises than we expected but its release is two or more years away
(at least after Jan. 2016), so remove Release-Go1.3Maybe tag from this issue for now.

Labels changed: removed release-go1.3.

Status changed to New.

rsc · 2014-05-21T14:21:02Z

Comment 10:

If we need a portable frexp/ldexp I'll write a much shorter one than the SunPro code you
found.

Labels changed: added release-go1.4.

Status changed to Accepted.

rsc · 2014-09-25T18:28:43Z

Comment 11:

Can assign a release once the release date is clearer. #9 says it is at least after Jan
2016

Labels changed: added release-none, removed release-go1.4.

Owner changed to @minux.

l-d-x · 2015-08-20T00:30:14Z

types_freebsd.go if_data8 needs to be changed to work with FreeBSD-11 for network interfaces. see my comments in issue #11641

mikioh · 2015-08-20T20:15:10Z

@l-d-x,

Thanks for the information of ABI changes in 11-CURRENT.

mpasternacki · 2015-08-26T16:19:28Z

FreeBSD ports issue for if_data8: https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=202504

Right now, there is a conditionally applied patch that changes the syscall module on FreeBSD-11. I'm not sure what would be a proper approach to get this solved properly in Go source itself without losing compatibility with FreeBSD 10 and earlier. Would a C preprocessor conditional in types_freebsd.go be a good idea? In that case, what about the generated ztypes files that are checked into the repository?

Looking at the source, it seems like golang.org/x/sys/unix is also affected; not sure how to proceed on that front either.

ianlancetaylor · 2015-08-26T17:24:59Z

A C preprocessor conditional would not be a good idea.

I don't understand how the change in the if_data8 struct is causing trouble. As far as I can see, if_data8 is used in exactly one place: a field in if_msghdr8. And as far as I can see, if_msghdr8 is not used at all. I don't know why those types are defined.

I think the real problem is the type IfData in ztypes_freebsd_.go, and its effect on IfMsghdr, and its use in route_freebsd_.go. The code in route_freebsd_32bit.go already does crazy stuff based on the kernel version. We would just have to do more of it there and in route_freebsd_64bit.go. Probably we would have to change IfMsghdr to the FreeBSD11 version, which would break compatibility slightly but that's OK since this is the syscall package.

l-d-x · 2015-08-26T17:44:40Z

It's the offset and size of the ifi_datalen field that has changed in FreeBSD-11 which is causing problems; it's now 16-bits but types_freebsd/ztypes:IfData.Datalen is still 8bits. When parseInterfaceMessage accesses p.Header.Data.Datalen, it's not fetching the entire 16-bit value (little-endian results in a zero value from the MSB) so it returns a wrong InterfaceMessage.Data.

mikioh · 2015-08-26T17:45:04Z

already does crazy stuff ...

And freebsd/386 on freebsd/amd64 shows us more crazy things. I think the solution would be just adding support for if_msghdrl into x/sys/unix and internal/syscall packages. It provides TLV-style data access view (at last) so we can make a type like type IfMsghdrl struct { FixedHeaderFields; Len int; Data []byte } for freebsd10 and beyond without worrying about future ABI changes.

mikioh · 2016-03-09T00:18:58Z

It's a time to fix this issue. My rough plan is like the following:

net: pull routing message/socket parsers off from syscall and put them into a single internal package such as net/internal/route not only for fixing the current ABI breakages on FreeBSD 11 but supporting for future ABI/KBI changes,
os: looks like no change seems to be needed,
- freebsd-11/386 on freebsd-11/amd64 might be needed to test
runtime: the use of kqueue/kevent in network poller might require small changes,
- freebsd-11/386 on freebsd-11/amd64 might be needed to test
syscall: deprecate routing message APIs,
buildbot farm: having freebsd-11/amd64 buildbot is nice.
- freebsd-11/386 on freebsd-11/amd64 too

gopherbot · 2016-04-26T05:00:44Z

CL https://golang.org/cl/22451 mentions this issue.

gopherbot · 2016-05-10T09:00:40Z

CL https://golang.org/cl/22990 mentions this issue.

This change reorganizes test cases for surveying network interfaces and address prefixes to make sure which part of the functionality is broken. Updates golang#7849. Change-Id: If6918075802eef69a7f1ee040010b3c46f4f4b97 Reviewed-on: https://go-review.googlesource.com/22990 Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>

gopherbot added accepted labels Sep 25, 2014

rsc added this to the Unplanned milestone Apr 10, 2015

rsc removed release-none labels Apr 10, 2015

mikioh mentioned this issue Jul 9, 2015

net: TestInterfaces fails on FreeBSD 11-CURRENT #11641

Closed

mikioh added the OS-FreeBSD label Jul 9, 2015

mikioh modified the milestones: Go1.7, Unplanned Mar 9, 2016

mikioh mentioned this issue Mar 9, 2016

net: tips all.bash test failed: interface_test.go:74: route ip+net: invalid network interface name #14706

Closed

mikioh mentioned this issue Mar 9, 2016

proposal: x/net/route, vendor/golang.org/x/net/route: new package #14724

Closed

tajacobsen mentioned this issue Mar 11, 2016

Unexpected fault when doing drive pull on FreeBSD/arm (on Rapsberry Pi) odeke-em/drive#596

Open

mikioh mentioned this issue Apr 12, 2016

net: TestInterfaces fails on freebsd-arm #15262

Closed

gopherbot closed this as completed in b4bf066 May 15, 2016

golang locked and limited conversation to collaborators May 15, 2017

gopherbot added the FrozenDueToAge label May 15, 2017

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

all: add support for FreeBSD 11 #7849

all: add support for FreeBSD 11 #7849

gopherbot commented Apr 23, 2014

ianlancetaylor commented Apr 23, 2014

minux commented Apr 23, 2014

minux commented Apr 23, 2014

gopherbot commented Apr 24, 2014

minux commented Apr 24, 2014

gopherbot commented Apr 25, 2014

gopherbot commented Apr 25, 2014

minux commented Apr 25, 2014

mikioh commented Apr 25, 2014

rsc commented May 21, 2014

rsc commented Sep 25, 2014

l-d-x commented Aug 20, 2015

mikioh commented Aug 20, 2015

mpasternacki commented Aug 26, 2015

ianlancetaylor commented Aug 26, 2015

l-d-x commented Aug 26, 2015

mikioh commented Aug 26, 2015

mikioh commented Mar 9, 2016

gopherbot commented Apr 26, 2016

gopherbot commented May 10, 2016

all: add support for FreeBSD 11 #7849

all: add support for FreeBSD 11 #7849

Comments

gopherbot commented Apr 23, 2014

ianlancetaylor commented Apr 23, 2014

minux commented Apr 23, 2014

minux commented Apr 23, 2014

gopherbot commented Apr 24, 2014

minux commented Apr 24, 2014

gopherbot commented Apr 25, 2014

gopherbot commented Apr 25, 2014

minux commented Apr 25, 2014

mikioh commented Apr 25, 2014

rsc commented May 21, 2014

rsc commented Sep 25, 2014

l-d-x commented Aug 20, 2015

mikioh commented Aug 20, 2015

mpasternacki commented Aug 26, 2015

ianlancetaylor commented Aug 26, 2015

l-d-x commented Aug 26, 2015

mikioh commented Aug 26, 2015

mikioh commented Mar 9, 2016

gopherbot commented Apr 26, 2016

gopherbot commented May 10, 2016