Age | Commit message (Collapse) | Author |
|
Always encode with 32-bit creation value.
This reverts revert-commit a99df168b6d7ce37cfff781347ca8a15aaab133b.
|
|
This reverts revert-commit 1ba8bf12eddb363481b1b5856af7c4d7c3313e52.
|
|
|
|
After calculating the length of a list, length/1 would accidentally
bump all remaining reductions, forcing the process to yield. Correct
the calculation.
|
|
* maint:
Fix bad merge from maint-21
Fix bad merge from maint-21
Updated OTP version
Prepare release
# Conflicts:
# OTP_VERSION
# make/otp_version_tickets_in_merge
|
|
|
|
* maint-21:
Updated OTP version
Prepare release
|
|
|
|
* sverker/process_info-reductions-fix/OTP-15793:
erts: Force process_info(reductions) as signal
erts: Fix another bug in process_info(reductions)
|
|
|
|
* sverker/process_info-reductions-fix/OTP-15793:
erts: Force process_info(reductions) as signal
erts: Fix another bug in process_info(reductions)
|
|
Add hostname to -remsh if none is given
|
|
Not 100% sure this is needed to get correct reductions
as the direct query is not done if process is RUNNING anyway.
|
|
* lukas/erts/distr-frag-fixes:
erts: Add documentation for distribution fragments
|
|
* lukas/OTP-22/misc-fixes/OTP-15791:
erts: Fix so that non-oneshot pollset on OpenBSD works
erts: Remove dead cpu info code in erlexec
erts: Remove -instr from erlexec flags
|
|
The poll code for kevent on OpenBSD and NetBSD had not been fixed
to work properly when disabling events in the scheduler poll-set.
Because they use ONESHOT the delete operation was built into the
trigger and thus a lot of code was not needed when doing oneshot,
however it is needed when doing non-oneshot.
|
|
* sverker/tw-assert:
erts: Fix bug in timer wheel at cancel during bump yield
erts: Tidy up some harmless code typos
|
|
This was previously used when erlexec needed to decide
whether to run the smp or non-smp emulator.
|
|
|
|
* lukas/OTP-22/misc-fixes/OTP-15788:
erts: Fix PGO configure test for gcc
|
|
|
|
|
|
* lukas/erts/distr-frag-fixes:
Update run-dialyzer script to be more generic and aggressive
erts: Fix memory leak for down/exit message
|
|
The problem was that some on some platforms -fprofile-generate
does not imply -lgcov which means that it will not
link.
|
|
* lukas/OTP-22/misc-fixes:
erts: Restart driver_SUITE:polling node at failure
|
|
|
|
The net module uses socket, but when the system has
been built with --disable-esock, there is no socket...
|
|
|
|
Forgot about using 'otp_build update_preloaded --no-commit'
when updating the preloaded and compiled with plain erlc...
Caused problems for dyalizer...
OTP-15658
|
|
OTP-15658
|
|
The config options --[en|dis]able-esock now works as
expected.
|
|
The optional socket module is now not included in
the app file if esock is not enabled.
OTP-15658
|
|
Preliminary work to make socket configurable (enable and disable).
OTP-15658
|
|
* john/erts/alloc-header-packing-win64:
erts_alloc: Fix header packing on Win64
|
|
When a fragmented down/exit is sent to a non-existing process the message
would just be dropped without being deallocated.
|
|
Symptom:
Failed debug assert in find_next_timeout
ERTS_TW_ASSERT(tiw->yield_slot == ERTS_TW_SLOT_INACTIVE);
Problem:
If remove_timer() was called in between yielding erts_bump_timers()
tiw->true_next_timeout_time could be set to 0 leading to
find_next_timeout() being called before all bumping is done.
Solution:
Don't clear tiw->true_next_timeout_time in remove_timer()
if tiw->yield_slot is active.
Does not seem this bug could cause other more harmful symptoms,
but not sure.
|
|
|
|
The -name option already computes a default
hostname if none is given. This PR adds the
same behaviour to -remsh. Now we can run:
erl -name foo -remsh bar
erl -sname foo -remsh bar
This simplifies deployment scripts as otherwise
they have to compute the hostname by hand or
start an Erlang VM instance only to do so.
|
|
* rickard/dist-q-sz-fix:
Fix distribution queue size adjustment
|
|
|
|
|
|
|
|
to disregard consumed virtual reductions
which are not aggregated into p->reds
and if included may cause reduction count go backwards.
|
|
|
|
No need to pre-allocate a buffer before name length calculation.
|
|
where
$argc does not exists
"if $undefined_variable" evaluates as true !?
|
|
|
|
This reverts commit 425889ba69b69a9b6fb14bfbef121d51b78e853a.
|
|
|
|
|