Use lttng_ust_tracef instead of lttng_ust__tracef

[christophebedard]

Follow-up to #95
Follow-up to #98

The "official" symbol that the LTTng documentation mentions is:

1. >=2.13: lttng_ust_tracef(), not lttng_ust__tracef()
   1. Docs: https://lttng.org/docs/v2.13/#doc-tracef
   2. Code: https://github.com/lttng/lttng-ust/blob/2e8e974039487cb8b2b22dc01ffdd6c9931ec02b/include/lttng/tracef.h#L23
2. <=2.12: tracef(), not _lttng_ust_tracef()
   1. Docs: https://lttng.org/docs/v2.12/#doc-tracef
   2. Code: https://github.com/lttng/lttng-ust/blob/6b0c6b76480d8162cb00919367cc2d13ccbe7037/include/lttng/tracef.h#L35

I thought that the difference between them is that the former checks if the tracepoint is enabled before triggering the tracepoint, while the latter just triggers it, but that is not the case. We can observe this by not enabling lttng_ust_tracef:event in tracetools_test/case.py: the tracepoint does not get triggered. The latter still ends up checking if the tracepoint is enabled before triggering it. The difference is that the latter formats the string using vasprintf() and then tries to trigger the tracepoint, which may or may not trigger depending on whether it is enabled or not: https://github.com/lttng/lttng-ust/blob/9261aeadea4674c3d68e05c2e8fcbcefa54feb56/src/lib/lttng-ust/tracef.c#L46. Therefore, the former avoids formatting the string if the tracepoint is not enabled.

Anyway, all of this is to say that we should just use lttng_ust_tracef()/tracef() even though it's just a test.

This should be tested on RHEL because it uses LTTng <=2.12.

[clalancette]

Looks reasonable to me with green CI, including on RHEL.

[christophebedard]

• Linux
• Linux-aarch64
• Windows
• RHEL