IRC channel logs
2026-09-27.log
back to list of logs
<JohnCowan>Bindings in R6 libraries are immutable, and the same is true in R7 except that the REPL can change things: you can override an import wuth another import <TheTaoOfSu>The included URI libraries handle %-encoding, but I need to encode non-ASCII to punycode <old>ArneBab: in the REPL you should either do: ,option interp #t <old>or: ,option optimization-level 0 <old>then it does not matter if the modules are declarative or not <old>foor coop REPL I could not tell <old>I don't have the use case <old>I am 90% through making eval-in-sandbox having no continuation barrier <old>so that now we can evaluate in sanbox in a fibers thread <old>Not sure about time limit tho, <old>but you could in theory, for a website, allocate a limit on memory usage for executing a request <old>interresting, when aborting to a prompt, the unwind of a dynwind are executed on top of the current stack <old>that's very bad for call-with-stack-overflow-handler .. <old>that's a shame .. I want for it to be written in Scheme <old>but when the stack overflow, the unwind which is suppose to remove the overflow check from the VM will overflow itself .. <old>and what about all the unwind of the users? None of them can be run <old>Rn I disable the stack overflow when around the overflow handler so the handler can abort and all unwind can be run but then a sandbox application could basically do inifnite stack allocation in a unwind, thus escpaing the sandbox <old>I could instead bump the limit a little around the handle so unwind can run in a limited stack space when aborting <old>but then a user sandbox unwind could overflow and that would prevent my unwind to run that needs to pop the limit .. so the VM overflow handler stack would be in an inconsistan state <old>jeesh I can understand why this was implemented in C <ArneBab>old: does option interp #t work after the modules are already loaded? What I need is stepping into a module while the code runs (during the game loop or when requests are received) and replacing top level bindings from the REPL, so other procs in the module use the new version. <ArneBab>old: going down to optimization 0 would make hacking far more different from the production runtime, that’s why that’s something I’d want to avoid. <ArneBab>old: eval-in-sandbox in fibers sounds great! <mwette>ArneBab, you could wrap your top-level bound values in boxes or parameters.