overall, the i believe the issue of whitespace usage is a matter of
personal style, and should not be subject to strict rule enforcement; as
long as i can use CSOFF to disable rules on source files i create, then i
can accept rules which encode different usage patterns;
i believe it is more important to preserve the style of the original author
of a given source file rather than attempt to follow an arbitrary usage
pattern in this regard; i don't mind using rules that differ from mine when
those source files were authored by those different usage patterns; but i
do not agree with enforcing them in my own style for a variety of reasons:
- it slows me down
- it makes it harder for me to read my own code, since i am accustomed
to reading my style
ideally, i believe you should craft rules that are sufficiently flexible in
the area of personal style choices that accommodate all of our preferences;
however, if it is acceptable to deal with exceptions using CSOFF, etc.,
then that would be sufficient for me
On Mon, Feb 6, 2012 at 10:09 AM, Vincent Hennebert<[email protected]>wrote:
Hi Glenn,
Thanks for taking the time to look at this. Looks like we should be able
to reach a consensus without too much difficulty.
On 03/02/12 21:20, Glenn Adams wrote:
which version of checkstyle are you using? there are two errors in
parsing
the proposed checkstyle file with 5.1;
<!--<property name="ignoreEnhancedForColon" value="false"/> -->
<!--<module name="OneStatementPerLine"/> -->
once i fixed the checkstyle file to work with 5.1, i see that 4935 and
31935 new warnings/errors are introduced in trunk and in my i18n
branches,
respectively; clearly, this is going to require a large amount of editing
to allow use of the proposed rules;
Like I said most of them are purely about syntax and are easily solved
with a code formatting tool. Obviously I’m happy to run such a tool on
your own Git branches and submit a patch if that can help you.
i prefer not to use automatic tools to make fixups in this case for the
reasons that chris outlined
many of the new errors I notice (in both trunk and my i18n branches) have
to do with whitespace before or after '(', ')', and cast operations; i do
not agree with enforcing the presence or absence of whitespace around
these
constructs; i happen to always use whitespace before and after parens,
e.g., the following should produce no checkstyle warning:
public int foo ( int a, int b, int c ) {
return bar ( a, b, c );
};
I’d rather keep the rule, as it enforces standard Java style that will
be easily recognized by any Java developer. I also find the variation in
the use of whitespace to be one of the most distracting things when
reading code.
That said, if that really bothers you I would be ok with relaxing the
rule, except for the whitespace between a method call and the left
parenthesis, to make it clear that it’s a method call.
i prefer my style of using whitespace;
if you are not insisting that no CSOFF declarations appear in source code,
then I can accept your proposal, provided I can use CSOFF to disable this
rule for source files that i create (for those i didn't create, i can
adhere to the rule)
i would like whitespace after '{' and before '}' in an array
initialization, e.g., both of the following should be permitted:
int[] a = new int[] { 1, 2, 3 };
int[] a = new int[] {1, 2, 3};
Yep, no problem.
i would like SimplifyBooleanReturn to be removed;
Hmmm. Ok.
i would like whitespace after BNOT produce a warning, e.g. both ! foo and
!foo should be accepted without warning;
I’d keep the rule. Allowing a standalone exclamation point is too
dangerous I think. Too easy to miss.
again, if you don't mind me using CSOFF in source files I author, then I
can accept
i would like whitespace after DOT operator to be permissible, e.g., both
x.y and x . y should be permitted;
Why? Note that it’s possible to break the line before the dot.
when i cast an object reference then invoke a method, i like to use the
following whitespace:
( (Foo) obj ) . doit ( ... )
again, if you don't mind me using CSOFF in source files I author, then I
can accept
i would like empty blocks to be permissible, e.g., the following should
be
permitted:
if ( test ) {
/* TBD - handle test is true */
} else {
/* TBD - handle test is false */
}
I find that it’s just clutter, but I don’t really mind.
the issue is i would like to put comments into blocks that are otherwise
empty; this is useful as a reminder to me as a coder that i may need to do
something for those blocks in the future that make them non-empty
i would like the arbitrary line length rule to be removed; i do not agree
to 110 line length; or if you insist, i could accept 150;
I’m afraid I’m not happy to go any higher than 110. Pascal actually made
a good point by saying that there is only a certain number of tokens
that a human being can grasp on one line.
Also, having to scroll horizontally is a real pain and greatly impedes
the readability of code. And if the editor is set up to automatically
wrap long lines then this ruins the indentation, which is not any
better.
my screen width and editor (emacs) and font size allows me to use 150
characters without scrolling; i think this is just as reasonable as 110;
frankly both of these are arbitrary; rather than spending time arguing
about arbitrary limits, i would prefer simply removing this rule/limit, and
leaving it to peer review
i do not agree with including MultipleVariableDeclarations rule; i
routinely define multiple local variables in one statement, e.g., int x,
y;
I’d really rather keep the rule. Variables should be used (i.e.,
initialized) as soon as they are declared anyway. Otherwise there is
a risk that with time, the variable declaration appears further and
further away from where it is used.
i prefer being able to use multiple declarations in one statement when it
makes sense; the language allows it, so why should we impose an arbitrary
rule to prevent it?
again, if you don't mind me using CSOFF in source files I author, then I
can accept
i do not agree with requiring LocalFinalVariableName to match
'^[A-Z][A-Z0-9]*(_[A-Z0-9]+)*$';
instead, it should continue to match the currently used
pattern "^[a-z][a-zA-Z0-9]*$";
Unless I missed something this is the default in Checkstyle 5.5.
i got a warning under Checkstyle 5.1; so i guess the default has changed;
in order to avoid being subject to this change, i suggest it be made
explicit
why are there two NoWhitespaceAfter rules?
<module name="NoWhitespaceAfter">
<property name="tokens" value="ARRAY_INIT"/>
</module>
<module name="NoWhitespaceAfter">
<property name="allowLineBreaks" value="false"/>
<property name="tokens"
value="BNOT,DEC,DOT,INC,LNOT,UNARY_MINUS,UNARY_PLUS"/>
</module>
Because one rule allows a line break after the token, the other not. But
anyway, following your comment we would remove the former one.
ok
if you fix the above problems, then i will re-run on trunk and my i18n
branch to check if there are any other issues that need to be resolved;
Let me know what you think. Hopefully others will also speak up to share
their views.
Thanks,
Vincent
On Fri, Feb 3, 2012 at 10:45 AM, Vincent Hennebert<[email protected]
wrote:
Hi All,
it is well-known that people are not happy with the Checkstyle file we
have in FOP. And there’s no point enforcing the application of
Checkstyle rules if we don’t agree with them in the first place.
I’ve finally taken on me to create a new Checkstyle file that follows
modern development practices. I’ve been testing it on my own projects
for a few months now and I’m happy with it, so I’d like to share it with
the community. The idea is that once we’ve reached consensus on the
Checkstyle rules we want to apply, we could set up a no warning policy
and enforce it by running Checkstyle in CI.
I’m also taking this as an opportunity to propose that we adopt a common
Checkstyle policy to all the sub-projects of XML Graphics. So once we’ve
agreed on a set of rules we would apply them to FOP and XGC immediately,
and eventually also to Batik, and keep them in sync.
We would also apply the rules to the test files as well as the main
code. Tests are as important as the actual code and there is no reason
why they shouldn’t be checked.
It is likely that the current code will not be compliant with the new
rules. However, most of them are really just about the syntax, so
I believe it should be fairly straightforward to make the code at least
90% compliant just by applying Eclipse’s command-line code formatter.
Please find the Checkstyle file attached. It is based on Checkstyle 5.5
and basically follows Sun’s recommendations for Java styling with a few
adaptations. What’s noteworthy is the following:
• Removed checks for Javadoc. What we want is quality Javadoc, and that
is not something that Checkstyle can check. Having Javadoc checks is
counter-productive as it forces us to put {@inheritDoc} everywhere, or
to create truly useless doc like the following:
/**
* Returns the thing.
* @return the thing
*/
public Thing getThing() {
return thing;
}
This is just clutter really. I think it should be left to peer review
to check whether a Javadoc comment is properly written, or whether the
lack thereof is justified. There’s an excellent blog entry from
Torsten Curdt about this:
http://vafer.org/blog/20050323095453/
• Removed check for file and method lengths. I don’t think it makes
sense to define a maximum size for files and methods. Sometimes
a 10-line method is way too big, sometimes it makes sense to have it
reach 20 lines. Same for files: it’s ok to reach 1000 lines if the
class contains several inner classes. If it doesn’t, then it’s
probably too big. I don’t think there is any definite figure we can
agree on and blindly follow, so I think sizes should be left to peer
review.
• However, I left the check for maximum line length because unreasonably
long lines make the code hard to follow. I increased it to 110
though to follow the evolution of monitor sizes. But as Peter
suggested me, we probably want to keep it low in order to make
side-by-side comparison easy.
• I added a check for the order of imports; this is to reduce noise in
diffs when committing. I think most of us have configured their IDE to
automatically organise imports when saving changes to a file. This is
a great feature because it allows to keep the list of imports
up-to-date. But in order to avoid constant back and forth changes when
different committers change the same file, I think it makes sense that
we all have the same configuration. I modeled this list after
Jeremias’ one, that I progressively inferred from his commits.
Please let me know what you think. I’m inclined to follow lazy consensus
on this, and apply the proposed changes if nobody has objected within
2 weeks. If anybody feels that a formal vote is necessary, feel free to
say so.
Thanks,
Vincent
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
