| Commit message (Collapse) | Author | Age | Files | Lines |
|---|
| |\ |
|
| | | |
|
| | | |
|
| | | |
|
| |\ \ |
|
| | |\ \
| | |/ |
|
| | | | |
|
| |\ \ \
| |/ / |
|
| | |\ \
| | |/ |
|
| | | |\
| | | |
| | | |
| | | |
| | | | |
could discontinue this macro any moment, then we are prepared ....
As a bonus: time_t is now allowed to be 64-bit internally, without effect on the C API (like stub-enabled extensions)
|
| | | | |
| | | |
| | | |
| | | |
| | | | |
it, time_t will be 64-bit internally. But at API-level, time_t will still be restricted to 32-bit on Win32 (Not on Win64).
This keeps Tcl_StatBuf the same (unless USE_64BIT_TIME_T is defined), so 64-bit times still cannot be used everywhere.
|
| |\ \ \ \
| |/ / /
| | | |
| | | | |
Windows) without -Wunused-parameters warnings
|
| |\ \ \ \
| |/ / / |
|
| | |\ \ \
| | |/ /
| | | |
| | | | |
Update effective_tld_names.txt.gz to the latest version.
Adapt .gitattributes and .fossil-settings/binary-glob to handle *.gz file transparently in fossil/git
|
| | | |\ \
| | | |/ |
|
| | | | | |
|
| | |\ \ \
| | |/ / |
|
| | | | |
| | | |
| | | |
| | | |
| | | | |
Travis.
Fix a struct initializer, add two typecasts and a "const", which can generate gcc warnings with some compiler options.
|
| |/ / / |
|
| |\ \ \
| |/ / |
|
| | | |
| | |
| | |
| | | |
end-of-line spacing
|
| |\ \ \
| |/ /
| | |
| | | |
performance test suite)
|
| | |\ \
| | |/
| | |
| | | |
time-calibration cycle optimization [f6637d3dd8] (clock ticks never backwards, more precise and smooth drifting)
|
| | | |\
| | | |
| | | |
| | | | |
--baseline 0055a16a8b core-8-5-branch`)
|
| | | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | | |
NRT-capable):
- the clock ticks never backwards (avoid it by negative drifts using comparison of times before and after calibration);
- more precise, smooth/soft drifting (avoids too large drifts, already after 10 iterations the drift gets fewer as 0.1 microseconds);
- because of more accurate drifting (aspire to the smallest difference), we can prolong calibration interval (up to 10 seconds by small tdiff-value);
Closes ticket [b7b707a310ea42e9f1b29954ee8ca13ae91ccabe] "[win32] NRT-only - NativeGetTime backwards time-drifts bug"
|
| | | |\ \ |
|
| | |\ \ \ \
| | |_|/ /
| |/| | | |
|
| |\ \ \ \ \
| | |/ / /
| |/| | | |
|
| | |\ \ \ \
| | | |_|/
| | |/| | |
|
| | | | | | |
|
| |\ \ \ \ \
| |/ / / /
| | | | |
| | | | | |
deprecated in vc2017
|
| | |\ \ \ \
| | |/ / /
| | | | |
| | | | |
| | | | | |
deprecated in vc2017.
Also remove unix/config.status.lineno, which was -apparently - checked-in by accident in previous commit
|
| | | | |/
| | |/|
| | | |
| | | | |
deprecated in vc2017
|
| |\ \ \ \
| |/ / /
| | | |
| | | |
| | | | |
"sebres-8-6-clock-speedup-cr1" branch, but that will be merged again as soon as the work is done.
All other changes in "trunk" since then (e.g. the INST_STR_CONCAT1 performance improvement, and the removal of SunOS-4) are retained.
|
| |\ \ \ \
| | | | |
| | | | |
| | | | | |
performance-tests are kept)
|
| | | |_|/
| |/| |
| | | |
| | | | |
preventing a (char *) type case. No functional changes.
|
| | |\ \ \
| | |/ /
| |/| | |
|
| | |\ \ \
| | | |/
| | |/| |
|
| |\ \ \ \
| |_|_|/
|/| | | |
|
| | | | |
| | | |
| | | |
| | | | |
TclpGetDate/TclpLocaltime/TclpGmtime can be removed: They are internal and not used for anything any more.
|
| | | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | | |
denominator scale to/from microseconds, and therefore more precise "timerate" results under windows (using similar mechanisms as by Mac OSX).
Especially multi-threaded, because it works without lock opposite to microseconds (that use crictical section, because of the calibration thread).
The reason for usage of wide clicks instead microseconds explains following example (shows 20% performance deference):
% timerate -calibrate {}
% timerate {clock microseconds} 5000
0.297037 µs/# 14465901 # 3366585 #/sec 4296.906 nett-ms
% timerate {clock clicks} 5000
0.247797 µs/# 16869084 # 4035554 #/sec 4180.116 nett-ms
|
| |/ / /
| | |
| | |
| | |
| | |
| | |
| | | |
high resolution clicks are not microseconds based;
[win] use high resolution timer for the wide clicks and microseconds directly, prevent several forwards/backwards conversions;
[win, unix, mac-osx] normalize some functions for common usage in different time units (clicks, micro- and nanoseconds)
|
| |\ \ \
| | |/
| |/| |
|
| | |\ \
| | | |
| | | |
| | | |
| | | | |
core-8-5-branch (who said fossil doesn't have 'rebase'...), and moved the definition of "struct _timeb t" to the top of the function.
This branch is ready to be merged to core-8-5-branch, it looks good to me. Thanks to sebres (Dipl. Ing. Sergey G. Brester)
|
| | |/ /
|/| |
| | |
| | |
| | |
| | |
| | |
| | | |
counters actualized in calibration thread in UpdateTimeEachSecond;
This entails that sometimes sporadically time-drifts resp. jump-esque time-shifts occurred, what for example produces very confusing results during time measurement.
[unix] wrong cast fixed in TclpGetWideClicks: multiplication with 1000000 in long int may cause overflow
See ticket b87ad7e9146832d505f9a430d779c5313c440256
|
| | | | |
|
| | | |
| | |
| | |
| | | |
purpose, actually.
|
| | |/
|/| |
|
| |\ \
| |/
| |
| | |
Pietro Cerutti
|
| | |\
| | |
| | |
| | | |
Pietro Cerutti
|
