@Dmitri: I'm fine with checking in jwe's patch for now.  I will include some BIST tests and do that soon.

I just spent an hour on this and am still no closer to a solution.  There are several levels of function calls and the code is manipulating the stream state bits to communicate information in an out-of-band manner up and down the stack.  It would probably be wiser to actually have R/W parameters to the various functions so that we could easily distinguish between a true error on the stream and an EOF which should really just break out of various looping structures.

Possibly related bug https://savannah.gnu.org/bugs/?63467

Maybe the attached change?  It attempts to make the stream state handling for floating point conversions (performed in the octave_scan_conv<double> specialization) the same as what is used in the octave_scan_1 function that is used to process integer conversions.



(file #55762)

Bisected to:

changeset:   31542:2f08a53e0a23
parent:      31540:ce5b4a00b022
user:        Rik <rik@octave.org>
date:        Fri Nov 25 08:30:30 2022 -0800
summary:     Fix scanf handling of exceptional values (NA, NaN, Inf) at EOF (bug #63383)

31540 is good, 31542 is bad.

Dmitri.
--

This is also regression. I cannot reproduce this on octave 8.4 nor  7.3.

Dmitri.
--

This example shows a case where fscanf fails I think to correctly read an input:

np = 101;
x = linspace(0, 1, np)';
a = [x, x.^2];
dlmwrite('tst.dat', a, '\t');
fd = fopen('tst.dat', 'r');
ar = fscanf(fd, '%g %g', [2, Inf])';
fclose(fd);
ar(np - 5: end, :)
try
  [a(np - 5 : end, :), ar(np - 5 : end, :)]
catch me
  error('ran into %s', me.message);
end

octave output, featuring an extra row at the end:

ans =
   0.9500   0.9025
   0.9600   0.9216
   0.9700   0.9409
   0.9800   0.9604
   0.9900   0.9801
   1.0000   1.0000
   1.0000        0

error: ran into horizontal dimensions mismatch (6x2 vs 7x2)
error: called from
    tst at line 12 column 3

matlab output:

ans =
    0.9500    0.9025
    0.9600    0.9216
    0.9700    0.9409
    0.9800    0.9604
    0.9900    0.9801
    1.0000    1.0000

ans =
    0.9500    0.9025    0.9500    0.9025
    0.9600    0.9216    0.9600    0.9216
    0.9700    0.9409    0.9700    0.9409
    0.9800    0.9604    0.9800    0.9604
    0.9900    0.9801    0.9900    0.9801
    1.0000    1.0000    1.0000    1.0000