I was going to change int as index to long as index in RubyString but I realize that we will run into platform issues (which is likely why this was originally capped at MAX_INT). Without significant changes to how RubyString is defined (like using ropes/multiple lists internally) this will not get solved easily.

I am leaving this open as I think we can solve this but I do not anticipate us fixing this near term.

Is there another way to generate a massive string you'd recommend in the interim?

@mjc Perhaps make a string-like class which aggregates a list of strings. This may not sound appealing but it does have a useful property: Not all memory needs to be contiguously allocated. Finding a single contiguous region of 10G of memory is potentially challenging but 10 1G segments much less so.

@mjc out of curiosity, have you tried creating very large strings in MRI as well? I wonder if we're behaving appropriately here

Yes, this works fine in MRI, as big a string as memory and swap allow.

On Mon, Aug 17, 2015 at 10:57 AM, R. Tyler Croy notifications@github.com
wrote:

@mjc https://github.com/mjc out of curiosity, have you tried creating
very large strings in MRI as well? I wonder if we're behaving appropriately
here

—
Reply to this email directly or view it on GitHub
#3243 (comment).

@rtyler MRI is long so they may work or likely work for larger than max int

It's not possible for us to represent a large string like this with standard JVM code, since a byte[] in Java can only be indexed by a 32-bit signed integer. It would be possible to do this if we had a ByteList that used tricks like Unsafe, but that's a really big leap to make.

If we make aggregate bytelist with @nirvdrum bytelist extension changes we can support it through that but then we would need long signatures on bytelist (jotting notes here)