Difference between revisions of "EiffelStudio 7.0 Releases"

m (Bug fixes: Added a note for test#attach092.)
m (Bug fixes: Added a note for test#attach093.)
Line 45: Line 45:
 
* compiler: bug#17806 (test#attach090) - Taken into account attachment status of <e>like Current</e> when checking for conformance.
 
* compiler: bug#17806 (test#attach090) - Taken into account attachment status of <e>like Current</e> when checking for conformance.
 
* compiler: test#attach092 - Corrected processing of self-initializing attributes to follow the standard rules.
 
* compiler: test#attach092 - Corrected processing of self-initializing attributes to follow the standard rules.
 +
* compiler: test#attach093 - Ensured the bodies of self-initialized attributes are not considered as setting other attributes because their evaluation is optional and depends on whether the attributes were accessed before or not.
 
* compiler: test#scoop020 - Taken separateness status into account when checking for formal generic type equality and conformance as well as when substituting actual generics.
 
* compiler: test#scoop020 - Taken separateness status into account when checking for formal generic type equality and conformance as well as when substituting actual generics.
 
* base: bug#17806 - Made <e>{LINKED_STACK}.duplicate</e> void-safe.
 
* base: bug#17806 - Made <e>{LINKED_STACK}.duplicate</e> void-safe.

Revision as of 07:41, 16 September 2011

Below are not the release notes of EiffelStudio. It is the change log of EiffelStudio intermediate releases and the most important changes are highlighted in green or in red (for breaking changes). The release notes of EiffelStudio can be found at the following URL. You can also download the latest revision here. Information about other version can be found under Category:Releases.

EiffelStudio 7.0.x Releases

7.0.x

Placeholder for new stuff since last intermediate release.

New features

  • compiler: Supported development of incomplete void-safe classes (so called "design mode") by avoiding reporting void-safety errors for unreachable code, e.g. for the code after a call to a feature that never returns normally, say, to {EXCEPTIONS}.die. So a creation procedure that does not initialize attached attributes properly (because the corresponding effective classes are not available yet), can look like
make (...)
	do
		... -- Some attributes are not initialized.
		die (1) -- Compiler does not report VEVI errors.
	end
or
make (...)
	do
		... -- Some attributes are not initialized.
		check implemented: False then end -- Compiler does not report VEVI errors.
	end

Improvements

  • compiler: Enforced full class checking for void-safe classes.
  • compiler: Supported detection and report for the configuration error VD88 of void-safety mismatch when a descendant or a client have stricter void-safety setting than the corresponding ancestor or supplier.
  • compiler: Taken into account changes of class options when performing recompilation.
  • compiler: Added a new validity error VBAC(3) for assigner calls when the target is a stable query.
  • compiler: Supported an option "stable" for any feature, not only for an attribute.
  • compiler: Changed conditions under which VFAC(3) is reported to follow the recent modifications to the standard.

Feature removed

Bug fixes

  • compiler: bug#11675, bug#15010 - Included location information in VTCT error and VTCM warning reports.
  • compiler: bug#15178 (test#incr303) - Fixed bug that caused incorrect VTEC(2) report during incremental recompilation.
  • compiler: bug#16942 (test#anchor046), bug#17815 (test#anchor062) - Fixed a crash caused by a combination of a qualified anchored type and its dependency on a formal generic constraint.
  • compiler: bug#17731 (test#scoop013) - Fixed error in C code generation for separate feature calls when inlining is enabled.
  • compiler: bug#17736 (test#vffd802) - Avoided reporting VFFD(8) for process-relative once procedures.
  • compiler: bug#17764 (test#attach087) - Considered like Current as attached regardless of is_attached_by_default option.
  • compiler: bug#17765 (test#attach089) - Prohibited assignment of Void to a variable of a reference formal generic type that has no detachable mark.
  • compiler: bug#17766 (test#attach088) - Supported attachment marks on the type NONE, including implicit setting via attached-by-default option.
  • compiler: bug#17778 (test#attach062) - Corrected checks for attachment status of variables to include not only a body of a loop, but also its invariant, variant and exit condition when a variable may be detached in the loop body.
  • compiler: bug#17779 (test#attach091) - Corrected conditions for which VFAC(3) is reported to follow the forthcoming modifications to the standard. Supported detection and report of an error when VBAC(3) introduced in the forthcoming version is violated.
  • compiler: bug#17806 (test#attach090) - Taken into account attachment status of like Current when checking for conformance.
  • compiler: test#attach092 - Corrected processing of self-initializing attributes to follow the standard rules.
  • compiler: test#attach093 - Ensured the bodies of self-initialized attributes are not considered as setting other attributes because their evaluation is optional and depends on whether the attributes were accessed before or not.
  • compiler: test#scoop020 - Taken separateness status into account when checking for formal generic type equality and conformance as well as when substituting actual generics.
  • base: bug#17806 - Made {LINKED_STACK}.duplicate void-safe.
  • time: test#term195, test#eiffelstore001 - Made a postcondition in {INTERVAL}.intersection void-safe.
  • vision2: bug#17806 - Made {EV_TABLE}.duplicate and {EV_DYNAMIC_LIST}.new_chain void-safe.
  • vision2: bug#17824 - Avoided VUTA(2) validity rule violation in {EV_GRID_ROW_I}.update_parent_expanded_node_counts_recursively.

User changes

  • vision2: Relaxed precondition on `set_focus' to let you set the focus even if the widget is not displayed. Note that in this case the `set_focus' call will have no effect. This is necessary because testing Vision2 code in a window less environment could violate the precondition and it would be rather expensive to change all callers of `set_focus' to check this all the time.

Developer changes