nsfStack.c

Clone Tools
  • last updated 3 hours ago
Constraints
Constraints: committers
 
Constraints: files
Constraints: dates
- alias reform: instead of calling first an alias with a stack frame followed by the dispatched of the aliased cmd, resolve aliases internally in dispatch. This has the advantage that we do not have to ignore the "transparent" stack frame in some sitations, and we have much less problems with the names of the aliased cmds (e.g. objects) in the introspection methods. Additionally, several problem cases disappeared from the regression test cases. In addition, the new approach is faster. - eliminating obsolete flag NSF_CSC_CALL_IS_TRANSPARENT

  1. … 9 more files in changeset.
- generic/nsfStack.c, CscFinish(): Without further restricting object deletion based on activation counts, very-late method dispatches during the "physical" clean-up stage (e.g., a dispatch issued by an unset variable trace script) lead to premature object and class deletions, violating the total deletion order used in ObjectSystemsCleanup().

nsf.c, nsf.h, nsfStack.c, nx.tcl, tcl-cool.tcl, xotcl2.tcl - fix spelling in comments - strip unneeded space

  1. … 6 more files in changeset.
- update copyright notices with MIT licens from OSI (http://www.opensource.org/licenses/MIT)

  1. … 26 more files in changeset.
- provide more hints for gcc

  1. … 2 more files in changeset.
- minor optimization

- minor optimizations

  1. … 2 more files in changeset.
- minor optimizations

  1. … 1 more file in changeset.
* added support for calling private methods via -local and filters * extended regression test for private + filters

  1. … 6 more files in changeset.
* per-thread MEM_COUNT tested with aolserver/naviserver * removed INTERP macros for MEM_COUNT (since we use now per-thread tables instead of per-interp tables as in the first implementation)

  1. … 5 more files in changeset.
- nx.tcl * renamed "attribute" to "property" * renamed "defaultAttributeCallProtection" to "defaultPropertyCallProtection" * renamed "nx::Attribute" to ""nx::VariableSlot" * renamed "BootStrapAttributeSlot" to "BootStrapVariableSlot" * renamed "createBootstrapAttributeSlots" to "createBootstrapVariableSlots" * removed method attributes * implemented old attributes definition in xotcl2 as method "parameter"

  1. … 32 more files in changeset.
- nsf.c: * enabled MEM_COUNT debugging for multi-threaded apps. We collect the MEM_COUNT statistics now per interp. Therefore, we have to pass around "interp" in case where alloc() and free() or refCount functions are used (textually, a big change) * verified, that nsf runs clean in aolserver/naviserver (all INCR_REF_COUNTS all balanced)

  1. … 8 more files in changeset.
nsf.c: * provide debug-refcounts for "class.activationCount" * provide debug-refcounts for "object.activationCount" * deactivated CHECK_ACTIVATION_COUNTS oer default * tested refcounts with Tcl 8.6b2, found bug in Tcl 8.6b2 and submitted patch to sourceforge

  1. … 3 more files in changeset.
- nsf.c: * refined refcounting debugging * fixed various refcounting bugs, especially in error cases. * added explicit INCR_REF_COUNTs on Tcl_Objs with 0-refCount to ease debugging

  1. … 9 more files in changeset.
- generic/nsfStack.c / CallStackPopAll(): When cleaning up the unstacked CSC entries, the borderline case of just 1 CSC entry led to a dangling pointer (and so a crash). This was caused by the unhandled side effect of CscListRemove() (or, rather, NsfClassListUnlink()) which updates the interp's cscList pointer in the 1-list-entry scenario. By using a copy of this pointer in the cleanup loop, this scoped pointer became dangling. To reproduce the issue (prior to this commit), run for example:

Object create o { exit -1 }

I extended the CscListRemove() interface for a conditional out

parameter, to be used for tracking changes to the interp's cscList

pointer.

- generic/nsf.c: Modified the occurrences of CscListRemove() to

reflect its extended interface.

- generic/nsfStack.c: Extended the documentation snippets of the

CscList* operations accordingly.

  1. … 1 more file in changeset.
- nsf.c: improve performance (branch prediction) by using likely/unlikely macros for gcc

  1. … 3 more files in changeset.
- make sure to place declarations only at the begin of a block - minor code cleanup

  1. … 1 more file in changeset.
- generic/nsf.c: Fixed [current callingmethod] for submethods (i.e., reporting submethods as calling methods, including their methodpath). Added tests covering the fix (see tests/submethods.test). - generic/nsfStack.c (CallStackMethodPath): Re-arranged the traversal slightly, to escape non-ensemble frames (eval CFRAMEs) as parts of the method path. Added test (see tests/submethods.test). - library/nx/nx.tcl: Corrected the impl of Ensemble->defaultmethod() (e.g., used self instead of current) - tests/disposition.test: Fixed test cases turned out to be false positives. Before the fixes, [current methodpath] gave non-empty results for simple object aliases.

  1. … 4 more files in changeset.
- adding spaces after comma in argument lists

  1. … 5 more files in changeset.
- generic/nsf.c / NsfOConfigureMethod(): Configure adds frames to its callstack context, however, so far the callstack modifications have not honoured the fact that configure() can be called from uplevel'ed call sites. The callstack context resulting from a prior uplevel was erased by the call frame introductions. As a consequence, parameter-dispatched methods found themselves in a wrong variable frame context. The suggested patch adds uplevel awareness in the most generic way I could think of. The entire test suite runs against the patched NsfOConfigureMethod(). However, I might be missing unwanted consequences for untested cases ... we need to think of further test scenarios ...

  1. … 1 more file in changeset.
Squashed commit of the following:

commit 901efa4cb4e1bbd55a8c47ebb79f4c7b4b397668

Author: ssoberni <stefan.sobernig@wu-wien.ac.at>

Date: Wed Jul 6 17:51:38 2011 +0200

- generic/nsf.c: Provide the [current class] context to submethods

- tests/disposition.test: Added some tests on the interaction of

INACTIVE frames (mixin, filter) and submethods

- tests/submethods.test: Testing [current class] context for submethods

  1. … 3 more files in changeset.
- generic/nsfStack.c: Revised my former patch for traversing callstacks under ensemle dispatches. We fixed the traversal-ending gateway expression; and added transparency for INACTIVE frames when assembling the method path. Added further tests, remain to be completed ...

  1. … 2 more files in changeset.
Merge branch '2.0.0-develop' of ssh://alice.wu-wien.ac.at/usr/local/src/git-repo/xotcl into 2.0.0-develop

  1. … 1 more file in changeset.
- preliminary fix for volatile called through redsidual args - new regression test file volatile.test

  1. … 4 more files in changeset.
- nsfStack.c / CallStackMethodPath(): Providing a fix for reporting method paths correctly when dispatching to ensembles from alias parameters. Prior to that fix, the intermediate (and inactive) CMETHOD frame wrapping configure() calls caused [current methodpath]s to report empty string elements. Added some test cases. - nsf.c / ObjectDispatch(), DispatchUnknownMethod(): When testing objects as alias parameter targets, I found cases when the default (C-level) unknown handler was invoked upon. The result was an unknown message which reported the delegator object as "unknown" method! This was due to injecting the delegator object into the argument vector at position 0, subsequently used by the C-level unknown handler. First, I tried to make the C-level unknown handler aware of the ensemble call context (by funneling through the CSC and frame flags) but this turned out to be complicated to achieve this simple task. As an alternative, I suggest the following patch: The first argument in unknown calls is treated as a list, and dependening on the call context (ensemble vs. non-ensemble) contains all the necessary call data. For ensemble methods, it is the delegator object, the entire method path (as returned by [current methodpath]), and the actual unknown selector. The default C-level handler, treating the call info as list, resorts to reporting the last element of the list which is always the unknown selector. Custom (ensemble-level, application-level) unknown handler can wrap and further process the information as necessary. To demonstrate the usefulness, I rewrote EnsembleObject->unknown() accordingly. I checked for Tcl_Obj leaks. - Added some tests on objects as targets of alias object parameter. To be continued.

  1. … 6 more files in changeset.
* The newest trunk version of Tcl in fossil has TclStackFree() and TclStackAlloc() removed We had to substitute this functions. Unfortunately, the lifetime of the strack structures has changed, so we had shuffle some interals around.

  1. … 4 more files in changeset.
DTrace: * added "nsf::configure dtrace on|off" for skipping package initialization (to be handled in D script) * make compliation clean

  1. … 7 more files in changeset.
- dtrace improvements: * track objects in method invocations * trace result codes in method-return * add some dtrace sample scripts

  1. … 10 more files in changeset.
- first steps towards DTRACE support

  1. … 5 more files in changeset.
- cleanup and documentation of nsf-specific interp state - nsf::configure: added an option "profile on|off" (per default off)

  1. … 5 more files in changeset.