Para-byond-tracy glues together a byond server the tracy profiler allowing you to analyze and visualize proc calls. This differs from the standard byond-tracy by the nature of it writing to flatfiles inside data/profiler/
. While these files can be up to 30 gigabytes in size, they allow the RAM usage of DreamDaemon to remain very low at runtime, which is useful for profiling entire rounds (A profile from a full round needs >48GB of RAM just to view, let alone save at runtime).
Note that the files generated cannot be loaded straight into tracy. You must use replay.py
to load the .utracy
file and stream "it over the network" (localhost) into capture.exe
as part of Tracy. You can stream straight into Tracy.exe
, but this is not advised due to performance overhead.
The above script requires the lz4
library with the stream addon. The instructions for that are out of scope of this guide.
Update 2023-12-29: You can now use https://github.com/AffectedArc07/ParaTracyReplay to stream the files much faster than the python script.
Update 2024-04-18: You can now use https://github.com/Dimach/rtracy to stream the files even faster than the C# code, as well as split traces and other cool things.
A massive thanks to mafemergency
for even making this possible. The below readme is adapted from the original repo (branch: stream-to-file
) https://github.com/mafemergency/byond-tracy/
windows | linux |
---|---|
515.1647 | 515.1647 |
515.1646 | 515.1646 |
515.1645 | 515.1645 |
515.1644 | 515.1644 |
515.1643 | 515.1643 |
515.1642 | 515.1642 |
515.1641 | 515.1641 |
515.1640 | 515.1640 |
515.1639 | 515.1639 |
515.1638 | 515.1638 |
515.1637 | 515.1637 |
515.1637 | 515.1637 |
515.1636 | 515.1636 |
515.1635 | 515.1635 |
515.1634 | 515.1634 |
515.1633 | 515.1633 |
515.1632 | 515.1632 |
515.1631 | 515.1631 |
515.1630 | 515.1630 |
515.1627 | 515.1627 |
515.1626 | 515.1626 |
515.1625 | 515.1625 |
515.1624 | 515.1624 |
515.1623 | 515.1623 |
515.1622 | 515.1622 |
515.1621 | 515.1621 |
515.1620 | 515.1620 |
515.1619 | 515.1619 |
515.1618 | 515.1618 |
515.1617 | 515.1617 |
515.1616 | 515.1616 |
515.1615 | 515.1615 |
515.1614 | 515.1614 |
515.1613 | 515.1613 |
515.1612 | |
515.1611 | 515.1611 |
515.1610 | 515.1610 |
515.1609 | 515.1609 |
515.1608 | 515.1608 |
515.1607 | 515.1607 |
515.1606 | 515.1606 |
515.1605 | 515.1605 |
515.1604 | 515.1604 |
515.1603 | 515.1603 |
515.1602 | 515.1602 |
515.1601 | 515.1601 |
515.1600 | 515.1600 |
515.1599 | 515.1599 |
515.1598 | 515.1598 |
515.1597 | 515.1597 |
515.1596 | 515.1596 |
515.1595 | 515.1595 |
515.1594 | 515.1594 |
515.1593 | 515.1593 |
515.1592 | 515.1592 |
515.1591 | 515.1591 |
515.1590 | 515.1590 |
514.* | 514.* |
None built in, you need to reply these.
simply call init
from prof.dll
to begin writing to a file inside data/profiler
init will return the filename it writes to upon success - the filename will always start with .
// returns a string filename whenever
/proc/prof_init()
var/lib
switch(world.system_type)
if(MS_WINDOWS) lib = "prof.dll"
if(UNIX) lib = "libprof.so"
else CRASH("unsupported platform")
var/init = call_ext(lib, "init")()
if(length(init) != 0 && init[1] == ".") // if first character is ., then it returned the output filename
return init
else if("0" != init)
CRASH("[lib] init error: [init]")
/world/New()
var/utracy_filename = prof_init()
. = ..()
no build system included, simply invoke your preferred c11 compiler. examples:
(AA recommended: If you have the MSVC++ buildchain, open x86 Native Tools Command Prompt for VS 2022
and then cd to this repo. cl
should be on your path inside of that CLI environment)
(Absolucy recommended: just use clang on windows, tbh)
cl.exe /nologo /std:c11 /O2 /LD /DNDEBUG prof.c ws2_32.lib /Fe:prof.dll /experimental:c11atomics
clang.exe -std=c11 -m32 -shared -Ofast3 -DNDEBUG -fuse-ld=lld-link prof.c -lws2_32 -o prof.dll
gcc -D_FILE_OFFSET_BITS=64 -std=c11 -m32 -shared -fPIC -Ofast -s -DNDEBUG prof.c -pthread -o libprof.so
byond-tracy is in its infancy and is not production ready for live servers.