[This one contains no criticism, just explanations. With, again,
views from the command line world, and some problem resolution
at the end.]
Well done, Eric!
I am referring to the "git merge"s you recently did, the first one
from last Sunday being:
commit 4d69436d09a9a10544ca5b8cbf6280fc442ad7b6
Merge: 1b098dd 1a1008f
Author: Eric Iverson <eiver...@jsoftware.com>
Date: Sun May 22 13:38:48 2016 -0400
bpctlz
Followed up by the further four branch merges on Thursday.
This much better than your
commit 2135cfd8421c656afc961f7add4b2f0d7f9030d1
Author: Eric Iverson <eiver...@jsoftware.com>
Date: Wed May 11 16:39:03 2016 -0400
parser
where you integrated Henry's work by applying a large patch.
The downside of this patch is that it lumps the original, individual
commits together. (This is sometime wanted, but often not.) You
also lost all of Henry's original commit messages which had been:
Small instruction tweaks; fix error-message omitted when referring
to undefined x. etc. in explicit def
Add testcases and final tweaks to commentary
Reorder name-stacking to reduce testing and function-call time.
Renumber parser lines so that token-error info is in the line
number. Use locals to reduce pointer dereferencing. Total perf
improvement now 9.2%
Rewrite nvr routines for speed. Don't call nvrredef on unchanged
names. Much commentary. This brings the new parser's total
lint-on-lint improvement to 9.1%
Parser rewritten, with 6.7% speedup on lint-on-lint. The only
interface change is that the caller is no longer required to
allocate and initialize the parser stack.
and substituted them all by that single word "parser".
Merges (as you now did) preserve all this info while causing you
less work. Another huge advantage: Change dates and authorship
are preserved -- "git blame" will now associate buggy lines in
that merged code with Henry and not with you.
Because you have been doing merges like a pro, you certainly
need no further convincing. I just want to confirm that you
are doing things perfectly now.
Unless a merge involves significant integration work, it is
customary to just stick with the default merge commit message.
You already did that once before:
commit 33d0816a15117cb3e0af04b13a02d7e8b87a916f
Merge: 6c0777c 7a540ba
Author: Eric Iverson <eiver...@jsoftware.com>
Date: Tue Jan 19 11:14:20 2016 -0500
Merge branch 'master' of ssh://www.jsoftware.com:49198/jsource
There is one small problem remaing now, though:
Because you applied Henry's work from the "parser" branch by applying
it as a patch (instead of a merge), git isn't aware that these changes
are already present in the trunk.
Visually, one can see this in the github "network" view:
<https://gaertner.de/~neitzel/eref/parse-network.png>
or, equivalently, the git command which shows unmerged changes which
reside on topic branches only:
jsource 581 > git show-branch -a --topic | cut -c -72
* [master] simlify version template
! [origin/0quotecobignum] Leading zeros caused large numbers not to be
! [origin/Edotnonchar] Ensure adjusting the lower interval bound doesn
! [origin/HEAD] simlify version template
! [origin/bpctlz] Perf improvement: fix bsum() to use word-wide oper
! [origin/inplacealias] Added commentary. It is OK to overwrite th
! [origin/makevsdoc] First try at modifying jsoftware repository
! [origin/master] simlify version template
! [origin/oldbugs] Fix uninitialized variable encountered during
! [origin/parser] Small instruction tweaks; fix error-message o
! [origin/shapeandfill] Add more tests
! [origin/tallstackcrash] Fix so exec on fill-cell in " does
------------
+ [origin/inplacealias] Added commentary. It is OK to overwr
+ [origin/inplacealias^] Fix local-to-global aliasing during
+ [origin/oldbugs] Fix uninitialized variable encountered dur
+ [origin/oldbugs^] Back out change to prevent ^:gerund from
+ [origin/oldbugs~2] Fix bugs in explicit definition:
+ [origin/oldbugs~3] Fix old bugs:
+ [origin/parser] Small instruction tweaks; fix error-message
+ [origin/parser^] Add testcases and final tweaks to commenta
+ [origin/parser~2] Reorder name-stacking to reduce testing a
+ [origin/parser~3] Rewrite nvr routines for speed. Don't ca
+ [origin/parser~4] Parser rewritten, with 6.7% speedup on li
*+++++++++++ [origin/makevsdoc^] debug start command
jsource 582 >
The same thing with funky colors, and not eliding commits merged to
the master branch:
<https://gaertner.de/~neitzel/eref/parse-show-branch.png>
Now, how to resolve this?
First, we should check check manually that Eric applied all code from
Henry's the "parser" branch thruthfully. We can do this by recreating
the respective diffs and comparing them:
# First, Ericss single commit, diff against it's parent (= prior version).
# The commit id 2135cfd is taken from waaay earlier in this message:
jsource 605 > git diff 2135cfd^..2135cfd > /tmp/erics-single-diff
# find the place where Henry created his branch,
# and create the diff encompassing the entire branch:
jsource 606 > git merge-base origin/parser origin/master
725c4dcbd4d3cc01754d73aa27a2f6a68b70dda6
jsource 607 > git diff 725c4dcbd..origin/parser > /tmp/henry-branch-diff
# compare the two via checksumming:
jsource 608 > md5 /tmp/*-diff
MD5 (/tmp/erics-single-diff) = 5c14ece15d47c1652dd73a9982b20b4c
MD5 (/tmp/henry-branch-diff) = 5c14ece15d47c1652dd73a9982b20b4c
Verdict: it's the same.
At this point, Eric might be tempted to
1. undo the single patch ("git revert 2135cfd")
2. git merge parser
This would technically work but this form of "rewritig history" (the
entire history since Wed May 11) is not good for a public branch as
exposed as the master branch. One would need *everybody* to update
their git copies with the new history or a heap of confusion ensues
("what exactly was in v2.c on May, 19th?"). One does that kind of thing
(fixing history) on "mostly private" branches only. On publicly
branches, this is a big no-no and frowned upon. Within our current,
small J src community, it would be viable. You'd to give a big
shout "please everybody pull!" though, to avoid said confusion.
Luckily for Eric, things can be done much simpler: it is still possible
to do a "git merge" of the "parser" branch, just by doing the
git merge parser
At this point, the parser source code is already on the branch and nothing
in the code will be changed. It will also still be attributed to Eric,
not Henry:
bjsource 629 > git blame jsrc/p.c | grep token-in-error
2135cfd8 (Eric Iverson 2016-05-11 16:39:03 -0400 208) if(i&6){ // if not
PNOMATCH... (we test this way to start calculating the token-in-error offset)
2135cfd8 (Eric Iverson 2016-05-11 16:39:03 -0400 213) *dci =
(I)stack[(i&6)|1]; // set the token-in-error, using the offset encoded into i
BUT Henry's commits 0b4e216..c18f66a will now be be seen as integrated
into the master branch (I did it on my local master, of course):
jsource 632 > git log --oneline --graph | cut -c -55
* 71deab3 Merge remote-tracking branch 'origin/parser
|\
| * c18f66a Small instruction tweaks; fix error-message
| * 8b39f2f Add testcases and final tweaks to commentar
| * 386b7d6 Reorder name-stacking to reduce testing and
| * 9fe7c8a Rewrite nvr routines for speed. Don't call
| * 0b4e216 Parser rewritten, with 6.7% speedup on lint
* | edb3ec2 simlify version template
* | 19dee12 Henry - tallstackcrash
|\ \
| * | 696512e Fix so exec on fill-cell in " does not RE
| * | 4443998 Fix buffer overrun and ASSERT inside FPDE
* | | 86baf9c Henry - 0quotecobignum
|\ \ \
| * | | fcbb420 Leading zeros caused large numbers not
| * | | 2a4cb96 Make 0 ". bigvalue work even when bigva
| |/ /
[ ^ this line going WAY back, branching off from the Mon Apr 11 version.]
'Nuff for today,
Martin
----------------------------------------------------------------------
For information about J forums see http://www.jsoftware.com/forums.htm
