alignment faults. the code is:
0x00007f7ffd40ccd7 <_pixman_implementation_create_sse2+0>: push %rbp
0x00007f7ffd40ccd8 <_pixman_implementation_create_sse2+1>: mov %rsp,%rbp
0x00007f7ffd40ccdb <_pixman_implementation_create_sse2+4>: sub $0x10,%rsp
...
0x00007f7ffd40cdab <_pixman_implementation_create_sse2+212>: mov $0xffffffffffffffff,%rdx
0x00007f7ffd40cdb2 <_pixman_implementation_create_sse2+219>: mov %rdx,0xfffffffffffffff0(%rbp)
0x00007f7ffd40cdb6 <_pixman_implementation_create_sse2+223>: mov %rdx,0xfffffffffffffff8(%rbp)
0x00007f7ffd40cdba <_pixman_implementation_create_sse2+227>: movdqa 0xfffffffffffffff0(%rbp),%xmm0
upon entry we have stack aligned to 16 bytes. the first instruction
subtracts 8, we copy this not-16-byte aligned value into %rbp, and
then decrement the stack by 16 bytes.
then we store a big -1 into 16 bytes on the stack, and then load it
into %xmm0.. except we load it from an unaligned pointer and GP.
* syntax in 50-ypbind hook has been fixed
* man page corrections
* report hwaddr used by dhcpcd when debug is enabled
* Fix detecting inet address for INFORM support
* document reason RELEASE in dhcpcd-run-hooks
* Support RTM_CHGADDR in the upcoming NetBSD-6
This is used to work out if the hwaddr has changed as the interface
does not go down/up unlike other OSes
* ntp hook no longer attempts to restart ntpd if 1st attempt failed
- Postfix no longer automatically appends the system default CA
(certificate authority) certificates, when it reads the CA
certificates specified with {smtp, lmtp, smtpd}_tls_CAfile or
with {smtp, lmtp, smtpd}_tls_CApath. This prevents third-party
certificates from getting mail relay permission with the
permit_tls_all_clientcerts feature. Unfortunately, this change
may cause compatibility problems with configurations that rely
on certificate verification for other purposes. To get the old
behavior, specify "tls_append_default_CA = yes".
- A prior fix for compatibility with Postfix < 2.3 was incomplete.
When pipe-to-command delivery fails with a signal, mail is now
correctly deferred, instead of being returned to sender.
- Poor smtpd_proxy_filter TCP performance over loopback (127.0.0.1)
connections was fixed by adapting the output buffer size to the
MTU.
- The SMTP server no longer applies the reject_rhsbl_helo feature
to non-domain forms such as network addresses. This would cause
false positives with dbl.spamhaus.org.
- The Postfix SMTP server failed to deliver a "421" response and
hang up the connection after Milter error. Instead, the server
delivered a "503 Access denied" response and left the connection
open, due to some Postfix 1.1 workaround for RFC 2821.
- The milter_header_checks parser failed to enable any of the actions
that have no effect on message delivery (warn, replace, prepend,
ignore, dunno, and ok).
and the metadata required to interpret it. Callers of namei must now
create a pathbuf and pass it to NDINIT (instead of a string and a
uio_seg), then destroy the pathbuf after the namei session is
complete.
Update all namei call sites accordingly. Add a pathbuf(9) man page and
update namei(9).
The pathbuf interface also now appears in a couple of related
additional places that were passing string/uio_seg pairs that were
later fed into NDINIT. Update other call sites accordingly.
This reverts the default timeout for test cases back to 300 seconds.
The change in the release was quite blind because it did not anticipate
many existing tests to be slow enough to overflow the modified timeout
(30 seconds), specially in anita.
My plan to really fix this is to let test cases specify their sizes in
a declarative way instead of specifying timeouts in seconds (the timeout
being defined by atf-run on a size basis), so I'm not going to bother to
go over all existing tests trying to figure out which ones need a higher
timeout for now. It is just easier to revert.
interval instead of assuming that there are exactly 1000 real-time-clock
milliseconds per second! On some ports when running under qemu, there
can be twice as many RTC milliseconds as expected.
This is part 2 of the changes required to make the libevent tests work
on port-amd64 under qemu.
even when running under qemu on platforms with a clock-skew problem.
The original 3-second timer was intended to be "longer than the http
timeout" (which is 2 seconds), and the updated 5-second value still meets
this requirement. The updated value also meets the requirement even when
the http timeout stretches to 4-seconds under qemu.
This is part 1 of getting the libevent tests working on port-amd64 with
qemu.
Experimental version released on November 7th, 2010.
* Added the ATF_REQUIRE_THROW_RE to atf-c++, which is the same as
ATF_REQUIRE_THROW but allows checking for the validity of the exception's
error message by means of a regular expression.
* Added the ATF_REQUIRE_MATCH to atf-c++, which allows checking for a
regular expression match in a string.
* Changed the default timeout for test cases from 5 minutes to 30 seconds.
30 seconds is long enough for virtually all tests to complete, and 5
minutes is a way too long pause in a test suite where a single test case
stalls.
* Deprecated the use.fs property. While this seemed like a good idea in
the first place to impose more control on what test cases can do, it
turns out to be bad. First, use.fs=false prevents bogus test cases
from dumping core so after-the-fact debugging is harder. Second,
supporting use.fs adds a lot of unnecessary complexity. atf-run will
now ignore any value provided to use.fs and will allow test cases to
freely access the file system if they wish to.
* Added the atf_tc_get_config_var_as_{bool,long}{,_wd} functions to the atf-c
library. The 'text' module became private in 0.11 but was being used
externally to simplify the parsing of configuration variables.
* Made atf-run recognize the 'unprivileged-user' configuration variable
and automatically drop root privileges when a test case sets
require.user=unprivileged. Note that this is, by no means, done for
security purposes; this is just for user convenience; tests should, in
general, not be blindly run as root in the first place.
it allows libelf work on /dev/ksyms.
XXX the name of the flag is a bit confusing and i think it's better to rename
MALLOCED to DATA_MALLOCED or such. but i don't think it's worth increasing
the diff against the upstream for it.
- put xf86-input drivers first then xf86-video
- put x86 (i386/amd64) first then other ${MACHINE}s in alphabetical order
XXX1: is it better to use Makefile.${MACHINE} (or something else)?
XXX2: should all ports include xf86-video-wsfb unconditionally?
XXX3: should we have common definitions for MI PCI video drivers?
atf-run is not twice as large as before. This is a pull-up of
699284e5c0d0a375958293e578af4e02d68d1182.
(I don't think it's reasonable to intentionally cripple tests as I have
just done here. In the future I would like to only report the output of
failed test cases, which would allow us to undo this, but not there yet.)
Experimental version released on October 20th, 2010.
* The ATF_CHECK* macros in atf-c++ were renamed to ATF_REQUIRE* to match
their counterparts in atf-c.
* Clearly separated the modules in atf-c that are supposed to be public
from those that are implementation details. The header files for the
internal modules are not installed any more.
* Made the atf-check tool private. It is only required by atf-sh and being
public has the danger of causing confusion. Also, making it private
simplifies the public API of atf.
* Changed atf-sh to enable per-command error checking (set -e) by default.
This catches many cases in which a test case is broken but it is not
reported as such because execution continues.
* Fixed the XSTL and CSS stylesheets to support expected failures.
* Use dynamically sized buffers for reading kernel link events
* Use the active link address
* Added option to dump a lease to stdout
* TEST mode now works correctly if an old lease is NAKed
* routes with the gateway = leased ip are now treated as host routes
tre will be compiled without approx and wchar/mulibyte support to
only match the minimum requirement to replace our spencer regex.
This needs a lot of testing.
Only enabled when USE_LIBTRE is set to `yes'.
996f9c26e07a86607f373c8f0243d57329c11543
ef98529abaf16e40a2e684496bf3da8f9ff0d09c
These prevent atf-run from stalling/crashing when a subprocess of a test
case stays around after the test case itself exits.
Reported, and verified working, by pooka@.
- build glsl as a host tool
- add the glsl objects to libmesa
- add/remove new/deleted sources for various components
- adjust the libmesa/libGL builds to pull in the new glsl objects
- re-apply the BUILDSYMLINK hack for eval.c/pixel.c/pixelstore.c
- use glsl to build the slang headers on the fly
XXX: i couldn't get the python/xml generator to run outside of
XXX: the libxcb build tree, so i just copied these from there,
XXX: but ideally files/Makefile could know how to do this.
Symlink dereference fix for Linux broke the build there; corrected.
Fix issues on ancient FreeBSD, QNX, ancient NetBSD and Minix.
Improved reliability of hash function detection.
Catch and report unhandled exceptions before they propagate to C land
The C++ interface to run test cases goes like this:
1) C++ run function -> 2) C run function -> 3) C++ wrapper for
test case -> 4) test case head/body/cleanup
The previous code caught and reported unhandled exceptions in 1).
However, such approach does not seem to work everywhere. It fails,
for example, in NetBSD/i386 but works in NetBSD/amd64. I am not sure
which platform implementation is correct nor if there even _is_ a
defined behavior. No matter what, it feels wrong and clunky.
Move the last-resort exception catching to happen in 3) so that
exceptions don't propagate back to C.
Fixes the test-programs/result_test:result_exception test case in
NetBSD/i386 5.99.34.
