| Ticket UUID: | 2037611 | ||
| Title: | NRE: streamline Tcl's evaluation stacks? | ||
| Type: | Bug | Created on: | 2008-08-04 13:56:28 |
| Submitter: | msofer | Assigned to: | msofer |
| Subsystem: | 45. Parsing and Eval | Severity: | |
| Priority: | 7 High | Last modified: | 2010-05-28 06:38:40 |
| Status: | Open | Closed by: | |
| Resolution: | None | Closed on: | |
| Version: | obsolete: 8.6a1 | ||
| Description: | ||||
The different stacks that are kept by Tcl to manage the flow of evaluation include: 1. The C stack (much relieved since NRE) 2. The CallFrame stack 3. The Tcl evaluation stack in iPtr->execEnvPtr 4. The CmdFrame stack (TIP #280) 5. The stack of bottomPtr structs in TEBC 6. NRE's callback stack Each is managed separately in a subtle dance to maintain coordination. The whole thing looks a bit too complicated, it could use some study and maybe a redesign. In particular, 2-5 are related but not identical, maybe they could be merged. | ||||
| User Comments: | ||||
msofer added on 2008-08-10 22:16:46:
Logged In: YES user_id=148712 Originator: YES A big callFrameReform/procReform is called for: as is today, data is also redundant and causes strange bugs when it gets out of sync. A notable example is that numCompiledLocals can be found in Proc, CallFrame and LocalCache. The separation of concerns between Proc, CallFrame, ByteCode needs a careful redesign. Changes here might break extensions that touch/alloc Proc and/or CallFrame directly. These include Itcl, XOTcl, tbcload. msofer added on 2008-08-10 22:42:21:
Logged In: YES user_id=148712 Originator: YES Note that at least since varReform the procPtr in CallFrame is not really needed: local variables are inspected at runtime from the LocalCache and not the Proc, the procPtr is only needed at frame-pushing time to read procPtr->cmdPtr->nsPtr. [info frame] also reads procPtr->cmdPtr->nsPtr and procPtr->cmdPtr->clientData from the CmdFrame for "non-standard commands" (?). msofer added on 2009-06-13 20:32:49:
Note that the current management of the bottomPtr stack (and the confusing state of TEBC) stand in the way of making coroutines NRE-aware. msofer added on 2010-05-28 06:38:40:
That last comment is now outdated: coros are NRE-aware, and TEBC got a bit better in the meantime. | ||||
