Like Emscripten, you could manually set functions as safe, for better performance.

When making a Loop->Multiple from merged branches, could we output a switch statement instead?

Thank you for bringing the Relooper problem class to my attention (and indirectly the Stackifier problem class).
.
I had previously assumed that bytecode was "always" a 100% decompilable substitute for missing source code (100% identical, except temporary names), as is the case with Smalltalk decompiler.
.
What might be the reason for the problems of Relooper and Stackifier that are not present in the Smalltalk system?

Catch the stack overflow message and display one alerting the user to the stack size setting instead.

This will probably mean switching from a thread to a child process.

disassembling a story file threw up this message "Disassembling the storyfile...Stopping on unknown object type 4 at 41549699"

If given a debug file (inline or external) we should be able to at least get function names, if not more.

It would let us immediately disassemble the functions instead of the fragile code that currently parses the file, skipping other objects, and would also tell us how long each function is, so that we can find late branch targets. This might be the only way to get dynamic jumps to work.

Decompile strings AOT

Add support for compiling against Emglken

Add more specific reasons to the FunctionSafety enum, and add an option to output a file list each function and its safety status.

Adjacent branch instructions could be combined before identifying basic blocks, to simply the output and improve performance. However I'm not sure if Inform actually compiles many adjacent branch instructions?

Currently nodes that are not part of a loop scc are not pulled out (ie, made a next edge) unless they are not dominated by the loop header. This means that a series of loops becomes deeply nested. As most loops have only one exit node, we can put it out as long as there isn't already a next edge. We could simply pull the first non-scc node out, or try to weigh the options somehow.

The most common Glk functions (those in perform_glk) could be directly output. Match on the function ID and number of arguments so that wrong argument number calls will fall through to the default implementation.

Optimise stack operands. Easy to do within a basic block, but across block boundaries would probably require some sort of phi function?