Skip to content

Commit 4bc6502

Browse files
chriscoolchriscool
committed
rn-51: add article about dropping 'extern' keyword
1 parent 8418ada commit 4bc6502

File tree

1 file changed

+103
-2
lines changed

1 file changed

+103
-2
lines changed

‎rev_news/drafts/edition-51.md‎

Lines changed: 103 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -49,9 +49,110 @@ This edition covers what happened during the month of April 2019.
4949
attendees to explore Munich and get to know each other before the
5050
unconference begins".
5151

52-
<!---
5352
### Reviews
54-
-->
53+
54+
*[[RFC PATCH]*.h: remove extern from function declarations](https://public-inbox.org/git/3e3b9d6c8e54b326809b95adabd3be39d2d0770f.1555111963.git.liu.denton@gmail.com/)
55+
56+
Denton Liu sent a big RFC patch that removes "extern" keywords in
57+
front of function declarations and that affects in big number of
58+
'.h' header files.
59+
60+
He said that he used the following command to remove the keywords:
61+
62+
```
63+
$ git ls-files '*.h' | xargs sed -i'' -e 's/^\(\s*\)extern \(.*(\)/\1\2/'
64+
```
65+
66+
And the following one to check the result:
67+
68+
```
69+
$ git grep -w extern '*.h'
70+
```
71+
72+
Jeff King, alias Peff, replied that dropping "extern" should not be
73+
done on function pointer declarations though, as a function pointer
74+
is a variable, so its visibility would change if "extern" is removed
75+
in front of it.
76+
77+
Junio agreed with Peff and explained that the reason we avoid
78+
wholesale updates "is because of an unnecessary bug like this one,
79+
not the one-time conflict resolution load that can be subcontracted
80+
out to "rerere" once dealt with ;-)".
81+
82+
Junio also suggested using "a tool more suited for the job
83+
(e.g. spatch)" as it "raises the confidence level of the end result
84+
a lot more than a "sed and then eyeball" rewrite", though we then
85+
have to be careful about possible "bugs in the instructions we give
86+
to the tool".
87+
88+
Denton then sent a [second version](https://public-inbox.org/git/[email protected]/)
89+
of his work. It was now a 3 patch long patch series.
90+
91+
The first patch used the `spatch` tool to do most of the "extern"
92+
keyword removals. The second patch still used `sed` to do it where
93+
`spatch` didn't work, but it made sure to ignore function
94+
variables. The third patch used a Coccinelle script to make sure
95+
that no more "extern" keywords are added to the code base.
96+
97+
Thomas Gummerer replied to the first patch that it "leaves a bunch
98+
of oddly indented parameter lists behind", as when a function has a
99+
larger number of parameters, its declaration often span more than
100+
one line.
101+
102+
Denton then send a [third version](https://public-inbox.org/git/[email protected]/)
103+
of his work, with an additional patch to "manually align parameter
104+
lists that were mangled by the tools".
105+
106+
Peff replied wondering if it was a good thing to modify files in the
107+
compat/ directory as the code there is often "copied from
108+
elsewhere". He also wondered if it was worth keeping the Coccinelle
109+
script as it is expensive to run.
110+
111+
In the [fourth version](https://public-inbox.org/git/[email protected]/)
112+
of his work, Denton decided to keep the Coccinelle script, but
113+
excluded the compat/ directory from the files where "extern" is
114+
removed.
115+
116+
Junio noticed that in a commit message Denton mentioned using:
117+
118+
```
119+
$ git ls-files \*.{c,h} | \
120+
grep -v ^compat/ | \
121+
xargs sed -i'' -e 's/^\(\s*\)extern \([^(]*([^*]\)/\1\2/'
122+
```
123+
124+
and then asked him to remove the useless backslashes at the end of
125+
the first 2 lines above, saying:
126+
127+
> Your shell knows its own syntax sufficiently well, and when you end
128+
> a line with a pipe, it knows that you haven't finished speaking to
129+
> it and waits for you to give the command downstream of the pipe,
130+
> without being told that you haven't finished with a backslash.
131+
132+
Unfortunately SZEDER Gábor then found that the Coccinelle script
133+
failed on 'compat/obstack.h' as it is included in a '.c' file
134+
checked by Coccinelle scripts.
135+
136+
Denton replied by proposing to drop the patch that contained his
137+
Coccinelle script, but Johannes Schindelin, alias Dscho, suggested
138+
to exclude certain directories or certain files for certain
139+
Coccinelle scripts.
140+
141+
There was then a long discussion thread involving Peff, Denton,
142+
Jacob Keller, Dscho, Junio and Gábor about what was possible to do
143+
with Coccinelle and spatch and how they are used in the Git
144+
toolchain. This thread eventually resulted in Peff making a small
145+
series of
146+
[patches](https://public-inbox.org/git/[email protected]/),
147+
based on a previous patch by Jacob, that makes it possible to run
148+
Coccinelle faster on machines with a lot of memory. Peff's patches
149+
have since been merged into the 'master' branch.
150+
151+
Denton in the meantime sent a [fifth version](https://public-inbox.org/git/[email protected]/)
152+
of his work that drops the patch that contained his Coccinelle
153+
script, and loses the useless backslashes. This patch series has
154+
also been merged into the 'master' branch and will be part of Git
155+
v2.22.0 that should be released at the beginning of June 2019.
55156

56157
<!---
57158
### Support

0 commit comments

Comments
(0)