Bug #473
closedCan't trace long double payload field?
100%
Description
Once more using lttng-ust/doc/examples/easy-ust/
, I have:
sample.c
[...] #include <unistd.h> #include <complex.h> //for _Complex types #define TRACEPOINT_DEFINE #include "sample_component_provider.h" int main(int argc, char **argv) { int i = 0; for (i = 0; i < 2000; i++) { tracepoint(sample_component, message, "Hello World", 1.E+2f, 1.E+3, 1.E+4L); usleep(10); } return 0; }
sample_component_provider.h
[...] TRACEPOINT_EVENT( sample_component, message, TP_ARGS(char *, text, float, floatvalue, double, doublevalue, long double, longdoublevalue), TP_FIELDS( ctf_string(message, text) ctf_float(float, floatfield, floatvalue) ctf_float(double, doublefield, doublevalue) ctf_float(long double, longdoublefield, longdoublevalue) ) ) [...]
Now I try to create a trace:
$ lttng create mylocalsession Session mylocalsession created. Traces will be written in /home/daniel/lttng-traces/mylocalsession-20130312-121313 $ lttng enable-event -u --all All UST events are enabled in channel channel0 $ lttng start Tracing started for session mylocalsession $ lttng destroy Session mylocalsession destroyed
Between lttng start
and lttng destroy
I run ./sample
from another console.
I get no error messages whatsovever, but the resulting trace has what seems like a correct metadata
file...but an empty channel0_0
file (probably not truly empty, just closed incorrectly).
The same exercise, omitting the long double
value, works fine.
Files
Updated by Mathieu Desnoyers over 11 years ago
Please attach your metadata file to the bug.
--
Mathieu Desnoyers
EfficiOS Inc.
http://www.efficios.com
Updated by Daniel U. Thibault over 11 years ago
Attached.
Updated by Daniel U. Thibault over 11 years ago
- File channel0_0 channel0_0 added
- File metadata metadata added
I run into similar trouble with double _Complex
payload fields:
TP_ARGS(char *, text, float, floatvalue, double, doublevalue, double _Complex, doublecomplexvalue), TP_FIELDS( ctf_string(message, text) ctf_float(float, floatfield, floatvalue) ctf_float(double, doublefield, doublevalue)
#include <unistd.h> #include <complex.h> //for _Complex types #define TRACEPOINT_DEFINE #include "sample_component_provider.h" int main(int argc, char **argv) { int i = 0; double _Complex dc = 6.2 + 7.3*I; for (i = 0; i < 2000; i++) { tracepoint(sample_component, message, "Hello World", 1.2345E+2f, 1.23456E+3, dc);
$ babeltrace -v ~/lttng-traces/mydoublecomplex-20130312-140047/ -w ~/lttng-traces/mydoublecomplex_n_all_f_all -n all -f all [verbose] Verbose mode active. [verbose] Converting from directory(ies): [verbose] /home/daniel/lttng-traces/mydoublecomplex-20130312-140047/ [verbose] Converting from format: ctf <default> [verbose] Converting to directory: /home/daniel/lttng-traces/mydoublecomplex_n_all_f_all [verbose] Converting to format: text <default> [verbose] CTF visitor: parent links creation... [verbose] done. [verbose] CTF visitor: semantic check... [verbose] done. [verbose] CTF visitor: metadata construction... [verbose] env.hostname = "edge-vb-u12" [verbose] env.domain = "ust" [verbose] env.tracer_name = "lttng-ust" [verbose] env.tracer_major = 2 [verbose] env.tracer_minor = 1 [verbose] env.tracer_patchlevel = 0 [verbose] env.vpid = 10421 [verbose] env.procname = "sample" [verbose] done. [error] Unexpected end of stream. Either the trace data stream is corrupted or metadata description does not match data layout. [error] Reading event failed. Error printing trace.
The trace files are attached. Note that a similar trial with float _Complex
did not crash but reports suspicious values for the field: the event wrote (5.0 + 3.1 i) but babeltrace reads "44.8".
Updated by Mathieu Desnoyers over 11 years ago
- Assignee set to David Goulet
When an event containing a float type is received by the sessiond, it should check the mant_dig value, and if it is 0, it should print an ERR and return an error to lttng-ust.
Updated by Daniel U. Thibault over 11 years ago
Not sure if this is related, but here goes. Using:
sample.c
#include <unistd.h> #include <complex.h> //for _Complex types #define TRACEPOINT_DEFINE #include "sample_component_provider.h" int main(int argc, char **argv) { int i = 0; for (i = 0; i < 2000; i++) { tracepoint(sample_component, message, "Hello World", 1.2345678E+38f, 1.2345678E+308); usleep(10); } return 0; }
sample_component_provider.h
[...] TRACEPOINT_EVENT( sample_component, message, TP_ARGS(char *, text, float, floatvalue, double, doublevalue), TP_FIELDS( ctf_string(message, text) ctf_float(float, floatfield, floatvalue) ctf_float(double, doublefield, doublevalue) ) ) [...]
If I then do:
$ lttng create mysingledouble Session mysingledouble created. Traces will be written in /home/daniel/lttng-traces/mysingledouble-20130312-170253 $ lttng enable-event -u --all All UST events are enabled in channel channel0 $ lttng start Tracing started for session mysingledouble $ lttng-sessiond: main.c:1258: send_socket_to_thread: Assertion `fd >= 0' failed.
I ran ./sample
after the lttng start
. Above, lttng-sessiond, lttng-consumerd
both die and the trace is lost. If I then repeat the procedure:
$ lttng create mysingledouble Session mysingledouble created. Traces will be written in /home/daniel/lttng-traces/mysingledouble-20130312-170714 $ lttng enable-event -u --all All UST events are enabled in channel channel0 $ lttng start Tracing started for session mysingledouble $ PERROR [11164/11170]: epoll_wait: Invalid argument (in compat_epoll_wait() at compat-epoll.c:229) $ lttng destroy Session mysingledouble destroyed
I get a normal trace. What is really suspicious is that the two outcomes alternate with perfect regularity if I continue trying to trace again and again.
Updated by David Goulet about 11 years ago
- Status changed from New to Confirmed
- Assignee deleted (
David Goulet) - Target version set to 2.2
Updated by Mathieu Desnoyers about 11 years ago
About the 3 bugs filed here (please file separately next time):
1) I'm preparing a patch to ensure sessiond refuses events that contain a long double field.
2) double _Complex is unsupported, so the behavior is undefined. Please file a feature request separately against lttng-ust if this feature request does not already exist.
3) the assert() in sessiond seems to be unrelated. Please file a separate bug report if you can reproduce against an updated lttng.
Updated by Mathieu Desnoyers about 11 years ago
- Status changed from Confirmed to Resolved
- % Done changed from 0 to 100
Applied in changeset 8494bda55cb1a09f5db0b57968fc3e84bd2981ea.
Updated by Daniel U. Thibault about 11 years ago
Mathieu Desnoyers wrote:
1) I'm preparing a patch to ensure sessiond refuses events that contain a long double field.
2) double _Complex is unsupported, so the behavior is undefined. Please file a feature request separately against lttng-ust if this feature request does not already exist.
3) the assert() in sessiond seems to be unrelated. Please file a separate bug report if you can reproduce against an updated lttng.
I'm unable to reproduce the assertion failure with LTTng 2.3.0-rc3. Must have been a fluke.
Long double fields compile in trace provider code and the event registration is rejected by sessiond. Is the intent that when/if lttng-sessiond accepts long double fields (once babeltrace does, I guess), the trace providers will start working without needing recompilation?