Checked field expressions, such as an object variant field access results in
occasionally broken address analysis crashing the VM. This guard added here
mimics guarded access in ldObjAddr as well. This is to prevent a crash, while a
fix is devised.
When statement branches exit early outside of nimvm. In nimvm it seems that all
sides of the branches must be evaluated as the code gen happens at a later
stage, this remains intact.
* Analyse last reads all at once
* Integrate firstWrite analysis
* Small cleanup
* Use sets instead of seqs
* Remove instrTargets
* Reap the benefits
* Implement error diagnostics
* Operate on DFA index for lastRead analysis
* Use mgetOrPut
* Cache alias results
This improves performance by a lot, since many
CFG locations map to a single PNode
* Improve performance
* Improve performance
* Cleanup
* Fix#17025
* Grammar
* Expand testcase
* basic stability improvements; refs nimsuggest
* fixed dot operator recursive loop & macro suggest
* hacky fix for run away dot operator sem check
Committing this mostly to make the issue more clear. Perhaps get better
feedback.
* semExprWithType seems like a better place to check
* fixed error messages const case expressions
* Clean-up test
* stopped the dot operator madness
No longer get infinite recursion when seming broken code with a dot
operator macro like in jsffi.
Co-authored-by: Araq <rumpf_a@web.de>
* Analyse last reads all at once
* Integrate firstWrite analysis
* Small cleanup
* Use sets instead of seqs
* Remove instrTargets
* Reap the benefits
* Implement error diagnostics
* Operate on DFA index for lastRead analysis
* Use mgetOrPut
* Cache alias results
This improves performance by a lot, since many
CFG locations map to a single PNode
* Improve performance
* Improve performance
* Cleanup
* Analyse last reads all at once
* Integrate firstWrite analysis
* Small cleanup
* Use sets instead of seqs
* Remove instrTargets
* Reap the benefits
* Implement error diagnostics
* Operate on DFA index for lastRead analysis
* Use mgetOrPut
* minor improvements
* IC: added the required logic for compilerProcs
* LazySym ftw
* we need this testing logic
* reimplement the old way we use for module package creation
* fixes a regression; don't pick module names if you can avoid it
* penalizes the quality score of deprecated symbols
* uses quality more pervasively in order to reflect deprecation impact
* impacts both sug and con
additional notes:
* linux i386 CI was failing
* this is because the suggested results differ slightly in their sort
* 64 bit tables.getOrDefault:441 was returned, while 32 bit returned 422
* for now simply removing the last line is good enough
