This fixes the bug introduced by #181, which was a bugfix for a bug reported in #180.
Before #181, when a non-zero off into the current input buffer was provided, the loop was conservative in what it would consume: rather than attempting to consume characters in the range [off, off + len), it would only consider the characters in the range [0, len).
181 changed this so that it considered the range [off, idx + len) where idx is the current position in the input buffer. This seemingly worked on simple parsers. However for more complex parsers this led to unsafe memory accesses when determining how much of the input to consume, and that resulted in a length that ran over the end of the input buffer. When an take_while* parser then tried to return the bytes to the user, a checked memory access would then fail.
This fixes the bug introduced by #181, which was a bugfix for a bug reported in #180.
Before #181, when a non-zero
offinto the current input buffer was provided, the loop was conservative in what it would consume: rather than attempting to consume characters in the range[off, off + len), it would only consider the characters in the range[0, len).181 changed this so that it considered the range
[off, idx + len)whereidxis the current position in the input buffer. This seemingly worked on simple parsers. However for more complex parsers this led to unsafe memory accesses when determining how much of the input to consume, and that resulted in a length that ran over the end of the input buffer. When antake_while*parser then tried to return the bytes to the user, a checked memory access would then fail.