[bug-diffutils] Re: Bug#577832: diffutils: [REGRESSION] newline is added to a line that has no newline if the line is in context (fwd)

[Paul Eggert]

Kai YU wrote:
> This repair does fix the bug. However,  this fix tries to adjust the
> final linbuf[line] setting rather than corrects the error cause.

Hmm, sorry, what is the "error cause" here?

> Instead, we suggest withdraw the change introduced in 2.9 may be a better
> choice.

We can't simply do that, because the change introduced for 2.9 in 2006
<http://git.savannah.gnu.org/cgit/diffutils.git/commit/?id=58d0483b621792959a485876aee05d799b6470de>
fixed a serious performance bug that was earlier introduced in 2001
<http://git.savannah.gnu.org/cgit/diffutils.git/commit/?id=420373575c3761cc50e762084400056affb968a0>.
The 2006 change simply reverts that (clearly incorrect) part of the
2001 change.  Perhaps there's a better fix than the combination of
the 2006 and later changes, but simply going back to the 2001
version would cause such a performance hit that we can't really do that.

> Besides, the value of filevec[0].suffix_begin in 3.0 is not equal to
> that in 2.8.1, which may cause potential bugs in the future.

Could you please give more details about the potential bugs that you
think might happen?  Sorry, I don't see them.

filevec[0].suffix_begin is supposed to be a pointer to the first line
of text that is in the common suffix (not counting the first few
lines of the suffix, as specified by the complicated expression
involving HORIZON_LINES).  Does the current code not do this
correctly in some cases?  Can you supply an example test case?