bufio() limits or bug?

[Mauricio Caceres]

Mata read/writes via bufio seem to be limited to under 2^31 bytes at a time (largest signed integer). Is this a bug or a limitation of reading/writing matrices? Apologies if I missed the docs for the latter. Simple example:

mata
C  = bufio()
fn = st_tempfilename()
fh = fopen(fn, "w")
fbufput(C, fh, "%8z", runiform(((2^31)/8)-1, 1))
fbufput(C, fh, "%8z", runiform(((2^31)/8),   1))
fclose(fh)
unlink(fn)
end

The second fbufput gives

                bufput():  3300  argument out of range
               fbufput():     -  function returned error
                 <istmt>:     -  function returned error
r(3300);

[Joseph Coveney]

Mata read/writes via bufio seem to be limited to under 2^31 bytes at a time . . . Is this a bug or a limitation . . . ? Apologies if I missed the docs for the latter.

It's the latter, and the documentation is in the Limits and memory utilization help file.

fbufput() is a convenience wrapper for bufput() for ease-of-use when writing to files.

The documentation for bufput() states, "bufput(C, B, offset, bfmt, X) aborts with error if B is too short to receive the result . . ."

B (the buffer) is string scalar, and in the limits help file linked to above, "string elements" (string scalars) in Mata are limited to 2,147,483,647 (i.e., 2^31-1) bytes.

[Mauricio Caceres]

Joseph Coveney Ah, I had seen the note you quote here on the limits for strings, but I figured those were strings, not numbers (which Stata suggests can be up to 2^48-1 in size).

What I hadn't clocked was the mechanic you pointed out that mata writes to B, and B is limited to 2^31-1 bytes. Thanks!