Bugzilla – Full Text Bug Listing
| Summary: | One issue related to privatedata in threads | ||
|---|---|---|---|
| Product: | [Tools] PTP | Reporter: | rong li <ronglli> |
| Component: | SCI | Assignee: | Tu Hong Jun <tuhongj> |
| Status: | CLOSED FIXED | QA Contact: | |
| Severity: | normal | ||
| Priority: | P3 | CC: | ronglli |
| Version: | unspecified | Keywords: | usability |
| Target Milestone: | --- | ||
| Hardware: | Other | ||
| OS: | Linux | ||
| Whiteboard: | |||
The fix has been checked in, Steve Cooper found this issue and provided most of the patch, thanks to him. Please close the bug if this issue is resolved. |
Build Identifier: 1.0 The latest version of SCI includes new code libsci/privatedata.{cpp,hpp}. This appears to be an attempt at allowing each thread to have its own 'private' data supported through the pthread library. The big problem that this forces a severe constraint on SCIs API. When SCI_Initialize() is called, the private data of the calling thread will be initialized. When the same thread then calls the API, say SCI_Bcast(), it works fine. However, if any other user thread makes the call its private data will not have been initialized. The SCI will then operate on a NULL pointer. Reproducible: Always