llvm-project/lldb/www/lldb-coding-conventions.html

140 lines
12 KiB
HTML

<!DOCTYPE html PUBLIC "-//W3C//DTD XHTML 1.0 Transitional//EN" "http://www.w3.org/TR/xhtml1/DTD/xhtml1-transitional.dtd">
<html xmlns="http://www.w3.org/1999/xhtml">
<head>
<meta http-equiv="Content-Type" content="text/html; charset=iso-8859-1" />
<link href="style.css" rel="stylesheet" type="text/css" />
<title>LLDB Tutorial</title>
</head>
<body>
<div class="www_title">
The <strong>LLDB</strong> Coding Conventions
</div>
<div id="container">
<div id="content">
<!--#include virtual="sidebar.incl"-->
<div id="middle">
<div class="post">
<h1 class ="postheader">LLDB Coding Conventions</h1>
<div class="postcontent">
<p>The LLDB coding conventions differ in a few important respects from LLVM.</p>
<p>
Note that <a href="http://clang.llvm.org/docs/ClangFormat.html">clang-format</a> will deal with
most of this for you, as such is suggested to run on patches before uploading. Note however that
clang-format is not smart enough to detect instances of humans intentionally trying to line variables
up on a particular column boundary, and it will reformat them to remove this "extraneous" whitespace.
While this is usually the correct behavior, LLDB does have many uses of manually aligned types and
fields, so please be aware of this behavior of clang-format when editing this type of code.
</p>
<p>
<b>Important</b>: Where not explicitly outlined below, assume that the
<a href="http://llvm.org/docs/CodingStandards.html">LLVM Coding Conventions</a> are to be followed.
</p>
<h3>Source code width:</h3>
<p>lldb does not follow the 80 character line restriction llvm imposes. In our
experience, trying to fit C++ code into an 80 character line results in code that
is awkward to read, and the time spent trying to find good indentation points to
avoid this would be much better spent on thinking about your code.
<p>More importantly, the restriction induces coders to choose overly abbreviated names
to make them better fit in 80 characters. In our opinion choosing good descriptive
names is much more important than fitting in 80 characters.
<p>In lldb the limit for code lines is 120 characters because it gets awkward to scan
longer lines even on a fairly big monitor, and we've found at that length you seldom
have to make code look ugly to get it to wrap.
<p>However you will see some instances of longer lines. The most common occurrence is in
the options tables for the CommandInterpreter, which contain the help strings as well as
a bunch of important but hard to remember fields. These tables are much easier to read if
all the fields line up vertically, and don't have help text interleaved in between the lines.
This is another thing to keep in mind when running clang-format, as it will always wrap at
120, so you will need to tweak its output when running against intentionally too-long lines.
<h3>Indentation:</h3>
<p>lldb uses 4 character indentation. We find this makes the code structure much easier to
see when scanning code, and since we aren't trying to fit code into 80 characters, the
benefit of not wasting 2 out of the 80 precious spaces per indentation level is moot.
<p>We also use the Allman brace style rather than putting the initial brace at the end
of the braced line. This makes the block structure of the code much easier to see on
an initial scan, and most folks have big enough monitors nowadays that saving a few
vertical lines isn't sufficiently important to outweigh this benefit.
<p>Though the llvm coding conventions don't specify this, llvm/clang tend to declare and
define methods by putting the return type and the method name on the same line. lldb
puts the qualifiers and return type on a line by themselves and then the method name on
the next line, i.e.:
<code><pre><tt>
virtual int
MethodName ();
</code></pre></tt>
<p>When you are scanning a header file, that makes the method names stand out more easily,
though at the cost of an extra line. When you have a editor that scrolls smoothly, it's
easy to move through pages so the extra line is less important than the ease of picking
out the method names, which is what you generally are scanning for.
<h3> Names:</h3>
<p>lldb's naming conventions are different and slightly more restrictive than the llvm
ones. The goal is to make it easy to tell from immediate context the lifespan
and what kind of entity a given name represents, which makes reading code you are not familiar
with much easier. lldb uses the following conventions:
<ul>
<li> Macro definitions when needed are in all caps, nothing else should be in all caps. </li>
<li>Types and classes are in CamelCase with an initial capital.</li>
<li>Methods are also in CamelCase with an initial capital. The initial capital for methods
has the handy benefit that it gets our method names into a different namespace
than the standard C/C++ library functions, which tend to all be lower-cased.
There are also places in lldb where we wrap clang objects in classes appropriate to lldb,
and the difference from the llvm convention here actually makes it easier to tell
whether you are using the clang object directly or are going through the lldb wrapper.</li>
<li> All variables are written in lower case, with "_" as the word separator. We find that
using a different capitalization and word separation convention makes variables and methods/types
immediately visually distinct, resulting in code which is much easier to read.</li>
<li> class ivars all start with "m_". It is important to be able to tell ivars from local
variables, and this makes the distinction easily apparent. Some other coding conventions
use an initial "_", but this seems much harder to spot. Also it allows:</li>
<li> Class statics and other global variables start with "g_". You should be suspicious of all
global variables, so having them stand out lexically is a good thing.</li>
<li>We also use the suffixes "_sp" and "_up" for shared and unique pointer variables. Since
these have very different lifecycle behaviors it is worthwhile to call them out
specially. You will see some "_ap" suffixes around. There should be no auto_ptr variables
left in lldb, but when we converted to unique_ptr's not all the names were changed.
Feel free to change these to "_up" when you touch them for some other reason.</li>
<li> enumerations that might end up being in the lldb SB API's should all be written like:
<pre><code><tt>
typedef enum EnumName
{
eEnumNameFirstValue,
eEnumNameSecondValue,
} EnumName;
</pre></code></tt>
<p>This redundancy is important because the enumerations that find their way through SWIG into
Python will show up as lldb.eEnumNameFirstValue, so including the enum name
in the value name disambiguates them in Python.
<p>Since we've started allowing C++11 in lldb, we have started using "enum class" instead of straight
enums. That is fine for enums that will only ever exist on the lldb_private side of lldb, but err on
the side of caution here on't do that for any enums that might find their way into the SB API's, since then
you will have to change them so we can get them through SWIG.</li>
<p> Also, on a more general note, except when you are using a temporary whose lifespan is not
far past its definition, never use one or two character names for ivars. Always use something
descriptive, and as far as possible use the same name for the same kind of thing (or the name
with an appropriate prefix.) That way if I'm looking at one use of a type, I can search on the
variable name and see most of the other uses of the same type of thing. That makes it much easier
to get quickly up to speed on how that type should be used.
</li>
</div>
</body>
</html>