Gnuplot versions through 6.0.5 and current 6.1
The initial fix for Bug #2896 showed a memory leak that is only tangentially related to the main issue in that bug report. Here is a reproducer that tests the problem more directly. At least part of it is that bad or missing data causes the column() function in a using specifier to return NaN and set an "undefined" flag. When df_readascii() sees this it terminates processing early and skips any string cleanup, so strings that were used in the expression are not freed.
There is also some weirdness going on that leads to the bad data itself being interpreted as a string.
Reproducer script attached, together with output from current 6.1 (bad) and a partial fix (no memory leak but still some unexpected string return).
Log in to post a comment.
