Re: [PATCH 0/2] add p in C tweaks

[Junio C Hamano <gitster@pobox.com>]

"Phillip Wood via GitGitGadget" <gitgitgadget@gmail.com> writes:

> A code cleanup and small bug fix for the C version of add -p
>
> dscho has pointed out that the bug fix in the second patch gets lost in the
> other changes and suggested adding the last member of the enum (which fixes
> the bug with handling 'e') as a separate patch. I'm unsure as it feels odd
> to split up the introduction of the flags - I'd be interested to hear what
> others think.

Essentially, the original was doing:

    - In early part of patch_update_file(), decide what option to show
      in s->buf using "if (undecided_previous >= 0)" etc. boolean
      expression that is tailored for each command;

    - In later part of patch_update_file(), after getting an answer
      to the end user, try to use the same boolean expression that
      is tailored for each command to see if the given command is
      acceptable.

and the bug was that each pair of boolean expressions that are
supposed to be identical were duplicated in two places, and one pair
was not identical by mistake.

Your [2/2] fixes it by turning the above to

    - In early part of patch_update_file(), decide what option to show
      in s->buf using "if (undecided_previous >= 0)" etc. boolean
      expression that is tailored for each command, *AND* record the
      fact that the command is allowed in the permitted bitmask.

    - In later part of patch_update_file(), after getting an answer
      to the end user, consult the permitted bitmask computed
      earlier to see if the given command is acceptable.

Since there no longer is duplicated boolean expressions that are
supposed to be the same but different by a bug, once this conversion
is made, it is impossible to have the bug.  For that reason, I do
not think the suggested split makes much sense.

A much saner split, if we have to split this step into two, would be
to first fix the bug keeping the code structure of the original,
i.e. the later part guards the 'e' command with

	if (hunk_index + 1 == file_diff->mode_change)

but the earlier part also required !file_diff->deleted, i.e. the
condition should have been

	if (hunk_index + 1 > file_diff->mode_change && !file_diff->deleted)

So without introducing enum and permitted bitmask, you can fix the
bug in place, replacing the incorrect boolean condition in the later
part that guards the 'e' command with a corrected one.  That would
be a minimum fix that can become your new [2/2], whose theme is "to
fix the bug with minumum change".

On top of that, you can convert the function again to reach the
final shape that writes each boolean condition only once and records
the permitted commands in the bitmask.  That can be your new [3/2],
whose these is "to make it impossible to introduce the bug previous
step fixed".


> Phillip Wood (2):
>   add -p: use ALLOC_GROW_BY instead of ALLOW_GROW
>   add -p: fix checking of user input
>
>  add-patch.c | 67 +++++++++++++++++++++++++++++++++--------------------
>  1 file changed, 42 insertions(+), 25 deletions(-)
>
>
> base-commit: 47ae905ffb98cc4d4fd90083da6bc8dab55d9ecc
> Published-As: https://github.com/gitgitgadget/git/releases/tag/pr-702%2Fphillipwood%2Fwip%2Fadd-p-fixes-v1
> Fetch-It-Via: git fetch https://github.com/gitgitgadget/git pr-702/phillipwood/wip/add-p-fixes-v1
> Pull-Request: https://github.com/gitgitgadget/git/pull/702