Changes in NSF 2.3.0

Dear Community, We are pleased to announce the availability of the Next Scripting Framework (NSF) 2.3.0. The changes are mostly perfective improvements in terms of bug fixes, API consistency, and backwards compatibility. See below for the details. Diff stats since 2.2.0: 189 files changed, 17743 insertions(+), 16889 deletions(-) (222 commits) Major changes relative to 2.2.0 are: New Features - NSF: * Reform of "uplevel" and "upvar" methods: - The methods "uplevel" and "upvar" (well as the underlying commands [current callinglevel] and [self callinglevel]) behave now more consistently and more similar to the XOTcl 1 semantics. This eases the migration of XOTcl 1 code. These methods are needed for "frame skipping", i.e. when methods using ":upvar" or ":uplevel" are overlaid with filters/mixin classes but should keep the uplevel semantics without these interceptors. Otherwise, adding a filter/mixin can easily break existing code. - The behavior of uplevel/upvar should be stable now with respect to Tcl's "namespace eval" and similar commands. - Error handling for providing potentially invalid level arguments was fixed. * Rename reform for "nsf::procs": NSF procs can now be safely renamed (or deleted) using the Tcl [rename] command. This was only partially supported before, and could also lead to crashes (when redefining a renamed NSF proc). * New object property "autonamed": NSF objects having been created using "new", rather than using "create", can now be tested for this condition using the object property "autonamed": % package req nx 2.3 % nsf::object::property [nx::Object new] autonamed 1 % nsf::object::property [nx::Object create o] autonamed 0 The property "autonamed" can be used to avoid having to rely on pattern matching of the command names. This is e.g. useful in the NaviServer/AOLserver blueprint management, where temporary objects should be omitted. * Extended object property "volatile": By setting or unsetting the object property "volatile", the volatility of a given NSF object changed dynamically. In previous releases, it was not possible to remove the volatility property of an object. % package req nx 2.3 % ::nx::Object create ::o -volatile ::o # query volatile property % nsf::object::property ::o volatile 1 # modify volatile property % nsf::object::property ::o volatile false 0 - NX: * Properties and variables now provide "exists" as an additional accessor method, in addition to "get": % package req nx 2.3 % nx::Class create C { :property -accessor public {a 1} } ::C % ::C create ::c1 ::c1 % if {[::c1 a exists]} { ::c1 a get } else { ::c1 a set "1" } ^^^^^^ This is mainly to provide for symmetry in the property and variable API, which otherwise requires one to resort to low-level tests for existence, e.g.: if {[::c1 eval {info exists :a}]} ... This breaks the property/ variable abstraction, because one cannot move to a different value store other than object variables w/o rewriting client code. * Fixed method combination for ensemble methods: In previous releases, calling [next] at the end of a next chain in ensemble methods could fire unintentionally the unknown handler. * nx::zip rewrite: nx::zip now uses the built-in Tcl 8.6 encode and decode facilities, if available, rather than requiring the Trf package to be present. - XOTcl: * Improved compatibility of XOTcl2 with XOTcl1 behavior for volatile objects (see "volatile reform" above). * Improved compatibility of XOTcl2 with XOTcl1 behavior for uplevel/upvar from within methods (see "uplevel/ upvar reform" above). - nx::serializer: * Improved backward compatibility: Preserve overriding accessor/ mutators methods (instprocs) for slots. - MongoDB: * Added JSON serializer: The "find all" and "bson" methods now provide for JSON-formatted results, on request. ::nx::mongo::Class "find all" -asJSON ::nx::mongo::Object bson asJSON This is to facilitate implementing single-page JavaScript applications, among others. * Fixed test suite; tested the NSF MongoDB binding against latest stable releases of MongoDB (4.0.9) and MongoDB-C driver (1.14.0). - Documentation: * Added documentation of uplevel and upvar methods. - Maintenance & bug fixes: * VLA reform: Avoid the use of variable-length arrays (VLA) when building argument vectors of Tcl_Objs internally to NSF. Use fixed-sized ones (stack-allocated for a size known at compile time or dynamically allocated above) to prevent from potential overflows and to produce more time-efficient instructions. * Tcl 8.7: Support for Tcl 8.7a1 and the upcoming, unreleased Tcl 8.7a2 ( branch "core-8-branch"). NSF compiles and its regression tests execute successfully (including TCL_NO_DEPRECATE). * Misc (esp. NSF/C): Ran valgrind checks, plugged one source of potential leaks (NsfProcStubDeleteProc) and one invalid read on a dangling pointer (NsfCCreateMethod). * Test suite: Fix recursion tests on 8.5 (Windows-only, forward.test) for platform-specific error messages (CheckCStack), obsolete for 8.6 with NRE. Guarded test case on recursive forwards, to avoid preemptive crashes on stack-size limited systems. - Build environments: * Microsoft C compilers: Turned off COMDAT folding (/opt:icf) under "nmake" builds which can lead to unusable, pointless function-pointer comparisons (Nsf_ConvertToSwitch vs. Nsf_ConvertToBoolean). * Improve robustness of configure/ make procedure in the absence of generated DTRACE artifacts so that they are truly conditional on the corresponding configure flags for DTRACE. * Improved robustness of Windows (nmake) installation routines by testing for the existence of the install directory. * Rendered inference of GIT commit in autotools setup more robust, in light of partial GIT checkouts (e.g., w/o tags). The detailed changelog is available at The Next Scripting Framework 2.3.0 (containing NX 2.3.0 and XOTcl 2.3.0) can be obtained from Please report issues and wishes by opening a ticket at Best regards - Gustaf Neumann - Stefan Sobernig