| Summary: | [LTTng] Latency analysis - tmf broadcast of event doesn't work | ||
|---|---|---|---|
| Product: | z_Archived | Reporter: | Mathieu Denis <mathieu.denis> |
| Component: | LinuxTools | Assignee: | Francois Chouinard <fchouinard> |
| Status: | CLOSED INVALID | QA Contact: | Francois Chouinard <fchouinard> |
| Severity: | normal | ||
| Priority: | P3 | CC: | mathieu.denis |
| Version: | unspecified | ||
| Target Milestone: | --- | ||
| Hardware: | PC | ||
| OS: | Linux | ||
| Whiteboard: | |||
| Bug Depends on: | 302489 | ||
| Bug Blocks: | 331467, 355141 | ||
|
Description
Mathieu Denis
I don't think that this has to do with inter-component communication... Historical recap: The original authors of the Latency Analysis/View opted for a scheme that required 2 passes over the trace. The first pass was used to collect some statistical data over the whole trace, while the second pass was used to properly populate the underlying data structure (and then the viewers). At the time, the authors didn't have a good one-pass solution for this problem and we agreed that 2 passes was acceptable as a temporary solution. It was understood that we would have to re-work this part before committing the work to Open Source. It is very likely that the first pass is performed (coalesced) at the same time as the Histogram View is built (and the Experiment is indexed, and the State System is built, and etc, etc, etc). I believe that the Latency View can be revisited to make it a 1-pass analysis. We can probably re-use some of the work that was done for the Histogram View where the view was decoupled from the model and the buckets are now managed dynamically (see https://bugs.eclipse.org/bugs/show_bug.cgi?id=352885). Moved to 355141 |