Issue #19452 has been updated by mame (Yusuke Endoh). Thank you all! kddeisz (Kevin Newton) wrote in #note-8:
In terms of memory consumption, I would think it would be 1-to-1 if we dropped node_id and replaced it with column
What concerns me slightly is that it makes it impossible to uniquely identify nodes that have the same column information, such as NODE_SCOPE or its child node. Perhaps it is not a very big deal for ErrorHighlight, though. Eregon (Benoit Daloze) wrote in #note-9:
I think we should measure, e.g. by how much % does it increase the memory of a Ruby program and a Rails app? I suspect it's not much higher than keeping the node id.
When adding node_id, I evaluated the memory usage with `rails s`. (See #17930.) It increased about 3 % (97 MB -> 100 MB). Not very small to ignore. ufuk (Ufuk Kayserilioglu) wrote in #note-11:
could we not store line/column information as varints
That makes sense. Although some efforts have already been made in compression, I guess that there would be room for improvement. ---------------------------------------- Feature #19452: `Thread::Backtrace::Location` should have column information if possible. https://bugs.ruby-lang.org/issues/19452#change-102326 * Author: ioquatix (Samuel Williams) * Status: Open * Priority: Normal ---------------------------------------- I discussed this with @mame and it would be pretty useful if we could also get the column information from exception backtrace location, even if it was slow. A POC: ```ruby class Thread::Backtrace::Location if defined?(RubyVM::AbstractSyntaxTree) def first_column RubyVM::AbstractSyntaxTree.of(self, keep_script_lines: true).first_column end else def first_column raise NotImplementedError end end end ``` It would be good to have a standard interface, so we follow the same interface as https://bugs.ruby-lang.org/issues/19451 and vice versa where it makes sense. I'll investigate it. -- https://bugs.ruby-lang.org/