Extending ctags with Regex parser (optlib)¶
| Maintainer: | Masatake YAMATO <yamato@redhat.com> |
|---|
Table of contents
- Regular expression (regex) engine
- Regex option argument flags
- Overriding the letter for file kind
- Generating fully qualified tags automatically from scope information
- Multi-line pattern match
- Advanced pattern matching with multiple regex tables
- Scheduling a guest parser with
_guestregex flag - Defining a subparser
- Translating an option file into C source code (optlib2c)
Exuberant Ctags allows a user to add a new parser to ctags with --langdef=<LANG>
and --regex-<LANG>=... options.
Universal Ctags follows and extends the design of Exuberant Ctags in more
powerful ways and call the feature as optlib parser, which is described in in
ctags-optlib(7) and the following sections.
ctags-optlib(7) is the primary document of the optlib parser feature. The following sections provide additional information and more advanced features. Note that some of the features are experimental, and will be marked as such in the documentation.
Lots of optlib parsers are included in Universal Ctags, optlib/*.ctags. They will be good examples when you develop your own parsers.
A optlib parser can be translated into C source code. Your optlib parser can thus easily become a built-in parser. See “Translating an option file into C source code (optlib2c)” for details.
Regular expression (regex) engine¶
Universal Ctags currently uses the POSIX Extended Regular Expressions (ERE) syntax as same as Exuberant Ctags.
During building Universal Ctags the configure script runs compatibility
tests of the regex engine in the system library. If tests pass the engine is
used, otherwise the regex engine imported from the GNU Gnulib library
is used. In the latter case, ctags --list-features will contain
gnulib_regex.
See regex(7) or the GNU Gnulib Manual
for the details of the regular expression syntax.
Note
The GNU regex engine supports some GNU extensions described here. Note that an optlib parser using the extensions may not work with Universal Ctags on some other systems.
The POSIX Extended Regular Expressions (ERE) does not support many of the “modern” extensions such as lazy captures, non-capturing grouping, atomic grouping, possessive quantifiers, look-ahead/behind, etc. It may be notoriously slow when backtracking.
A common error is forgetting that a
POSIX ERE engine is always greedy; the ‘*’ and ‘+’ quantifiers match
as much as possible, before backtracking from the end of their match.
For example this pattern:
foo.*bar
Will match this entire string, not just the first part:
foobar, bar, and even more bar
Another detail to keep in mind is how the regex engine treats newlines.
Universal Ctags compiles the regular expressions in the --regex-<LANG> and
--mline-regex-<LANG> options with REG_NEWLINE set. What that means is documented
in the
POSIX specification.
One obvious effect is that the regex special dot any-character ‘.’ does not match
newline characters, the ‘^’ anchor does match right after a newline, and
the ‘$’ anchor matches right before a newline. A more subtle issue is this text from the
chapter “Regular Expressions”;
“the use of literal <newline>s or any escape sequence equivalent produces undefined
results”. What that means is using a regex pattern with [^\n]+ is invalid,
and indeed in glibc produces very odd results. Never use ‘\n’ in patterns
for --regex-<LANG>, and never use them in non-matching bracket expressions
for --mline-regex-<LANG> patterns. For the experimental --_mtable-regex-<LANG>
you can safely use ‘\n’ because that regex is not compiled with REG_NEWLINE.
And it may also have some known “quirks”
with respect to escaping special characters in bracket expressions.
For example, a pattern of [^\]]+ is invalid in POSIX ERE, because the ‘]’ is
not special inside a bracket expression, and thus should not be escaped.
Most regex engines ignore this subtle detail in POSIX ERE, and instead allow
escaping it with ‘\]’ inside the bracket expression and treat it as the
literal character ‘]’. GNU glibc, however, does not generate an error but
instead considers it undefined behavior, and in fact it will match very odd
things. Instead you must use the more unintuitive [^]]+ syntax. The same
is technically true of other special characters inside a bracket expression,
such as [^\)]+, which should instead be [^)]+. The [^\)]+ will
appear to work usually, but only because what it is really doing is matching any
character but ‘\’ or ‘)’. The only exceptions for using ‘\’ inside a
bracket expression are for ‘\t’ and ‘\n’, which ctags converts to their
single literal character control codes before passing the pattern to glibc.
You should always test your regex patterns against test files with strings that do and do not match. Pay particular emphasis to when it should not match, and how much it matches when it should.
Regex option argument flags¶
Many regex-based options described in this document support additional arguments
in the form of long flags. Long flags are specified with surrounding ‘{’ and
‘}’.
The general format and placement is as follows:
Some examples:
Note that the last example only has two ‘/’ forward-slashes following
the regex pattern, as a shortened form when no kind-spec exists.
The --mline-regex-<LANG> option also follows the above format. The
experimental --_mtable-regex-<LANG> option follows a slightly
modified version as well.
Regex control flags¶
The regex matching can be controlled by adding flags to the --regex-<LANG>,
--mline-regex-<LANG>, and experimental --_mtable-regex-<LANG> options.
This is done by either using the single character short flags b, e and
i flags as explained in the ctags.1 man page, or by using long flags
described earlier. The long flags require more typing but are much more
readable.
The mapping between the older short flag names and long flag names is:
| short flag | long flag | description |
|---|---|---|
| b | basic | Posix basic regular expression syntax. |
| e | extend | Posix extended regular expression syntax (default). |
| i | icase | Case-insensitive matching. |
So the following --regex-<LANG> expression:
is the same as:
The characters ‘{’ and ‘}’ may not be suitable for command line
use, but long flags are mostly intended for option files.
Exclusive flag in regex¶
By default, lines read from the input files will be matched against all the
regular expressions defined with --regex-<LANG>. Each successfully matched
regular expression will emit a tag.
In some cases another policy, exclusive-matching, is preferable to the all-matching policy. Exclusive-matching means the rest of regular expressions are not tried if one of regular expressions is matched successfully, for that input line.
For specifying exclusive-matching the flags exclusive (long) and x
(short) were introduced. For example, this is used in
optlib/gdbinit.ctags for ignoring comment lines in gdb files,
as follows:
Comments in gdb files start with ‘#’ so the above line is the first regex
match line in gdbinit.ctags, so that subsequent regex matches are
not tried for the input line.
If an empty name pattern (//) is used for the --regex-<LANG> option,
ctags warns it as a wrong usage of the option. However, if the flags
exclusive or x is specified, the warning is suppressed.
This is useful to ignore matched patterns as above.
NOTE: This flag does not make sense in the multi-line --mline-regex-<LANG>
option nor the multi-table --_mtable-regex-<LANG> option.
Experimental flags¶
Note
These flags are experimental. They apply to all regex option
types: basic --regex-<LANG>, multi-line --mline-regex-<LANG>,
and the experimental multi-table --_mtable-regex-<LANG> option.
_extra
This flag indicates the tag should only be generated if the givenextratype is enabled, as explained in “Conditional tagging with extras”.
_field
This flag allows a regex match to add additional custom fields to the generated tag entry, as explained in “Adding custom fields to the tag output”.
_role
This flag allows a regex match to generate a reference tag entry and specify the role of the reference, as explained in “Capturing reference tags”.
_anonymous=PREFIX
This flag allows a regex match to generate an anonymous tag entry. ctags gives a name starting with
PREFIXand emits it. This flag is useful to record the position for a language object having no name. A lambda function in a functional programming language is a typical example of a language object having no name.Consider following input (
input.foo):(let ((f (lambda (x) (+ 1 x)))) ... )Consider following optlib file (
foo.ctags):You can get following tags file:
$ u-ctags --options=foo.ctags -o - /tmp/input.foo Le4679d360100 /tmp/input.foo /^(let ((f (lambda (x) (+ 1 x))))$/;" l
Conditional tagging with extras¶
If a matched pattern should only be tagged when an extra flag is enabled,
mark the pattern with {_extra=XNAME} where XNAME is the name of the
extra. You must define a XNAME with the
--_extradef-<LANG>=XNAME,DESCRIPTION option before defining a regex flag
marked {_extra=XNAME}.
if __name__ == '__main__':
do_something()
To capture the lines above in a python program (input.py), an extra flag can
be used.
The above optlib (python-main.ctags) introduces main extra to the Python parser.
The pattern matching is done only when the main is enabled.
$ ctags --options=python-main.ctags -o - --extras-Python='+{main}' input.py
__main__ input.py /^if __name__ == '__main__':$/;" f
Adding custom fields to the tag output¶
Exuberant Ctags allows just one of the specified groups in a regex pattern to be used as a part of the name of a tag entry.
Universal Ctags allows using the other groups in the regex pattern. An optlib parser can have its specific fields. The groups can be used as a value of the fields of a tag entry.
Let’s think about Unknown, an imaginary language.
Here is a source file (input.unknown) written in Unknown:
public func foo(n, m);
protected func bar(n);
private func baz(n,...);
With --regex-Unknown=... Exuberant Ctags can capture foo, bar, and baz
as names. Universal Ctags can attach extra context information to the
names as values for fields. Let’s focus on bar. protected is a
keyword to control how widely the identifier bar can be accessed.
(n) is the parameter list of bar. protected and (n) are
extra context information of bar.
With the following optlib file (unknown.ctags), ctags can attach
protected to the field protection and (n) to the field signature.
For the line protected func bar(n); you will get following tags output:
bar input.unknown /^protected func bar(n);$/;" f protection:protected signature:(n)
Let’s see the detail of unknown.ctags.
--_fielddef-<LANG>=name,description defines a new field for a parser
specified by <LANG>. Before defining a new field for the parser,
the parser must be defined with --langdef=<LANG>. protection is
the field name used in tags output. access scope is the description
used in the output of --list-fields and --list-fields=Unknown.
This defines a field named signature.
This option requests making a tag for the name that is specified with the group 3 of the
pattern, attaching the group 1 as a value for protection field to the tag, and attaching
the group 4 as a value for signature field to the tag. You can use the long regex flag
_field for attaching fields to a tag with the following notation rule:
{_field=FIELDNAME:GROUP}
--fields-<LANG>=[+|-]{FIELDNAME} can be used to enable or disable specified field.
When defining a new parser specific field, it is disabled by default. Enable the
field explicitly to use the field. See “Parser specific fields”
about --fields-<LANG> option.
passwd parser is a simple example that uses --fields-<LANG> option.
Capturing reference tags¶
To make a reference tag with an optlib parser, specify a role with
_role long regex flag. Let’s see an example:
A role must be defined before specifying it as value for _role flag.
--_roledef-<LANG>.<KIND>=<ROLE>,<ROLEDESC> option is for defining a role.
See the line, --regex-FOO=.... In this parser FOO, the name of an
imported module is captured as a reference tag with role imported.
For specifying <KIND> where the role is defined, you can use either a
kind letter or a kind name surrounded by ‘{’ and ‘}’.
The option has two parameters separated by a comma:
<ROLE>
the role name, and
<ROLEDESC>
the description of the role.
The first parameter is the name of the role. The role is defined in
the kind <KIND> of the language <LANG>. In the example,
imported role is defined in the module kind, which is specified
with m. You can use {module}, the name of the kind instead.
The kind specified in --_roledef-<LANG>.<KIND> option must be
defined before using the option. See the description of
--kinddef-<LANG> for defining a kind.
The roles are listed with --list-roles=<LANG>. The name and description
passed to --_roledef-<LANG>.<KIND> option are used in the output like:
$ ctags --langdef=FOO --kinddef-FOO=m,module,modules \
--_roledef-FOO.m='imported,imported module' --list-roles=FOO
#KIND(L/N) NAME ENABLED DESCRIPTION
m/module imported on imported module
If specifying _role regex flag multiple times with different roles, you can
assign multiple roles to a reference tag. See following input of C language
x = 0;
i += 1;
An ultra fine grained C parser may capture the variable x with
lvalue role and the variable i with lvalue and incremented
roles.
You can implement such roles by extending the built-in C parser:
$ ctags with --options=c-extra.ctags --extras=+r --fields=+r
i input.c /^i += 1;$/;" v roles:lvalue,incremented
x input.c /^x = 0;$/;" v roles:lvalue
Scope tracking in a regex parser¶
About the {scope=..} flag itself for scope tracking, see “FLAGS FOR
--regex-<LANG> OPTION” section of ctags-optlib(7).
Example 1:
# in /tmp/input.foo
class foo:
def bar(baz):
print(baz)
class goo:
def gar(gaz):
print(gaz)
$ ctags --options=/tmp/foo.ctags -o - /tmp/input.foo
bar /tmp/input.foo /^ def bar(baz):$/;" d class:foo
foo /tmp/input.foo /^class foo:$/;" c
gar /tmp/input.foo /^ def gar(gaz):$/;" d class:goo
goo /tmp/input.foo /^class goo:$/;" c
Example 2:
// in /tmp/input.pp
class foo {
int bar;
}
$ ctags --options=/tmp/pp.ctags -o - /tmp/input.pp
bar /tmp/input.pp /^ int bar$/;" v class:foo
foo /tmp/input.pp /^class foo {$/;" c
NOTE: This flag doesn’t work well with --mline-regex-<LANG>=.
Overriding the letter for file kind¶
One of the built-in tag kinds in Universal Ctags is the F file kind.
Overriding the letter for file kind is not allowed in Universal Ctags.
Warning
Don’t use F as a kind letter in your parser. (See issue #317 on github)
Generating fully qualified tags automatically from scope information¶
If scope fields are filled properly with {scope=...} regex flags,
you can use the field values for generating fully qualified tags.
About the {scope=..} flag itself, see “FLAGS FOR --regex-<LANG>
OPTION” section of ctags-optlib(7).
Specify {_autoFQTag} to the end of --langdef=<LANG> option like
--langdef=Foo{_autoFQTag} to make ctags generate fully qualified
tags automatically.
‘.’ is the (ctags global) default separator combining names into a
fully qualified tag. You can customize separators with
--_scopesep-<LANG>=... option.
input.foo:
class X
var y
end
foo.ctags:
Output:
$ u-ctags --quiet --options=./foo.ctags -o - input.foo
X input.foo /^class X$/;" c
y input.foo /^ var y$/;" v class:X
$ u-ctags --quiet --options=./foo.ctags --extras=+q -o - input.foo
X input.foo /^class X$/;" c
X.y input.foo /^ var y$/;" v class:X
y input.foo /^ var y$/;" v class:X
X.y is printed as a fully qualified tag when --extras=+q is given.
Customizing scope separators¶
Use --_scopesep-<LANG>=[<parent-kindLetter>]/<child-kindLetter>:<sep>
option for customizing if the language uses {_autoFQTag}.
parent-kindLetter
The kind letter for a tag of outer-scope.
You can use ‘
*’ for specifying as wildcards that means any kinds for a tag of outer-scope.If you omit
parent-kindLetter, the separator is used as a prefix for tags having the kind specified withchild-kindLetter. This prefix can be used to refer to global namespace or similar concepts if the language has one.
child-kindLetter
The kind letter for a tag of inner-scope.
You can use ‘
*’ for specifying as wildcards that means any kinds for a tag of inner-scope.
sep
In a qualified tag, if the outer-scope has kind andparent-kindLetterthe inner-scope haschild-kindLetter, thensepis instead in between the scope names in the generated tags file.
specifying ‘*’ as both parent-kindLetter and child-kindLetter
sets sep as the language default separator. It is used as fallback.
Specifying ‘*’ as child-kindLetter and omitting parent-kindLetter
sets sep as the language default prefix. It is used as fallback.
NOTE: There is no ctags global default prefix.
NOTE: _scopesep-<LANG>=... option affects only a parser that
enables _autoFQTag. A parser building full qualified tags
manually ignores the option.
Let’s see an example.
The input file is written in Tcl. Tcl parser is not an optlib
parser. However, it uses the _autoFQTag feature internally.
Therefore, _scopesep-Tcl= option works well. Tcl parser
defines two kinds n (namespace) and p (procedure).
By default, Tcl parser uses :: as scope separator. The parser also
uses :: as root prefix.
namespace eval N {
namespace eval M {
proc pr0 {s} {
puts $s
}
}
}
proc pr1 {s} {
puts $s
}
M is defined under the scope of N. pr0 is defined under the scope
of M. N and pr1 are at top level (so they are candidates to be added
prefixes). M and N are language objects with n (namespace) kind.
pr0 and pr1 are language objects with p (procedure) kind.
$ ctags -o - --extras=+q input.tcl
::N input.tcl /^namespace eval N {$/;" n
::N::M input.tcl /^ namespace eval M {$/;" n namespace:::N
::N::M::pr0 input.tcl /^ proc pr0 {s} {$/;" p namespace:::N::M
::pr1 input.tcl /^proc pr1 {s} {$/;" p
M input.tcl /^ namespace eval M {$/;" n namespace:::N
N input.tcl /^namespace eval N {$/;" n
pr0 input.tcl /^ proc pr0 {s} {$/;" p namespace:::N::M
pr1 input.tcl /^proc pr1 {s} {$/;" p
Let’s change the default separator to ->:
$ ctags -o - --extras=+q --_scopesep-Tcl='*/*:->' input.tcl
::N input.tcl /^namespace eval N {$/;" n
::N->M input.tcl /^ namespace eval M {$/;" n namespace:::N
::N->M->pr0 input.tcl /^ proc pr0 {s} {$/;" p namespace:::N->M
::pr1 input.tcl /^proc pr1 {s} {$/;" p
M input.tcl /^ namespace eval M {$/;" n namespace:::N
N input.tcl /^namespace eval N {$/;" n
pr0 input.tcl /^ proc pr0 {s} {$/;" p namespace:::N->M
pr1 input.tcl /^proc pr1 {s} {$/;" p
Let’s define ‘^’ as default prefix:
$ ctags -o - --extras=+q --_scopesep-Tcl='*/*:->' --_scopesep-Tcl='/*:^' input.tcl
M input.tcl /^ namespace eval M {$/;" n namespace:^N
N input.tcl /^namespace eval N {$/;" n
^N input.tcl /^namespace eval N {$/;" n
^N->M input.tcl /^ namespace eval M {$/;" n namespace:^N
^N->M->pr0 input.tcl /^ proc pr0 {s} {$/;" p namespace:^N->M
^pr1 input.tcl /^proc pr1 {s} {$/;" p
pr0 input.tcl /^ proc pr0 {s} {$/;" p namespace:^N->M
pr1 input.tcl /^proc pr1 {s} {$/;" p
Let’s override the specification of separator for combining a
namespace and a procedure with ‘+’: (About the separator for
combining a namespace and another namespace, ctags uses the default separator.)
$ ctags -o - --extras=+q --_scopesep-Tcl='*/*:->' --_scopesep-Tcl='/*:^' --_scopesep-Tcl='n/p:+' input.tcl
M input.tcl /^ namespace eval M {$/;" n namespace:^N
N input.tcl /^namespace eval N {$/;" n
^N input.tcl /^namespace eval N {$/;" n
^N->M input.tcl /^ namespace eval M {$/;" n namespace:^N
^N->M+pr0 input.tcl /^ proc pr0 {s} {$/;" p namespace:^N->M
^pr1 input.tcl /^proc pr1 {s} {$/;" p
pr0 input.tcl /^ proc pr0 {s} {$/;" p namespace:^N->M
pr1 input.tcl /^proc pr1 {s} {$/;" p
Let’s override the definition of prefix for a namespace with ‘@’:
(About the prefix for procedures, ctags uses the default prefix.)
$ ctags -o - --extras=+q --_scopesep-Tcl='*/*:->' --_scopesep-Tcl='/*:^' --_scopesep-Tcl='n/p:+' --_scopesep-Tcl='/n:@' input.tcl
@N input.tcl /^namespace eval N {$/;" n
@N->M input.tcl /^ namespace eval M {$/;" n namespace:@N
@N->M+pr0 input.tcl /^ proc pr0 {s} {$/;" p namespace:@N->M
M input.tcl /^ namespace eval M {$/;" n namespace:@N
N input.tcl /^namespace eval N {$/;" n
^pr1 input.tcl /^proc pr1 {s} {$/;" p
pr0 input.tcl /^ proc pr0 {s} {$/;" p namespace:@N->M
pr1 input.tcl /^proc pr1 {s} {$/;" p
Multi-line pattern match¶
We often need to scan multiple lines to generate a tag, whether due to needing contextual information to decide whether to tag or not, or to constrain generating tags to only certain cases, or to grab multiple substrings to generate the tag name.
Universal Ctags has two ways to accomplish this: multi-line regex options, and an experimental multi-table regex options described later.
The newly introduced --mline-regex-<LANG> is similar to --regex-<LANG>
except the pattern is applied to the whole file’s contents, not line by line.
This example is based on an issue #219 posted by @andreicristianpetcu:
// in input.java:
@Subscribe
public void catchEvent(SomeEvent e)
{
return;
}
@Subscribe
public void
recover(Exception e)
{
return;
}
The above java code is similar to the Java Spring
framework. The @Subscribe annotation is a keyword for the framework, and the
developer would like to have a tag generated for each method annotated with
@Subscribe, using the name of the method followed by a dash followed by the
type of the argument. For example the developer wants the tag name
Event-SomeEvent generated for the first method shown above.
To accomplish this, the developer creates a spring.ctags file with
the following:
And now using spring.ctags the tag file has this:
$ ctags -o - --options=./spring.ctags input.java
Event-SomeEvent input.java /^public void catchEvent(SomeEvent e)$/;" s line:2 language:javaspring
recover-Exception input.java /^ recover(Exception e)$/;" s line:10 language:javaspring
Multiline pattern flags¶
Note
These flags also apply to the experimental --_mtable-regex-<LANG>
option described later.
{mgroup=N}
This flag indicates the pattern should be applied to the whole file contents, not line by line.Nis the number of a capture group in the pattern, which is used to record the line number location of the tag. In the above example3is specified. The start position of the regex capture group 3, relative to the whole file is used.
Warning
You must add an {mgroup=N} flag to the multi-line
--mline-regex-<LANG> option, even if the N is 0 (meaning the
start position of the whole regex pattern). You do not need to add it for
the multi-table --_mtable-regex-<LANG>.
{_advanceTo=N[start|end]}
A regex pattern is applied to whole file’s contents iteratively. This long flag specifies from where the pattern should be applied in the next iteration for regex matching. When a pattern matches, the next pattern matching starts from the start or end of capture group
N. By default it advances to the end of the whole match (i.e.,{_advanceTo=0end}is the default).Let’s think about following input
def def abcConsider two sets of options,
foo.ctagsandbar.ctags.
foo.ctagsemits following tags output:def input.foo /^def def abc$/;" a
bar.ctagsemits following tags output:def input-0.bar /^def def abc$/;" a abc input-0.bar /^def def abc$/;" a
_advanceTo=1startis specified inbar.ctags. This allows ctags to captureabc.At the first iteration, the patterns of both
foo.ctagsandbar.ctagsmatch as follows0 1 (start) v v def def abc ^ 0,1 (end)
defat the group 1 is captured as a tag in both languages. At the next iteration, the positions where the pattern matching is applied to are not the same in the languages.
foo.ctags0end (default) v def def abc
bar.ctags1start (as specified in _advanceTo long flag) v def def abcThis difference of positions makes the difference of tags output.
A more relevant use-case is when
{_advanceTo=N[start|end]}is used in the experimental--_mtable-regex-<LANG>, to “advance” back to the beginning of a match, so that one can generate multiple tags for the same input line(s).
Note
This flag doesn’t work well with scope related flags and exclusive flags.
Advanced pattern matching with multiple regex tables¶
Note
This is a highly experimental feature. This will not go into the man page of 6.0. But let’s be honest, it’s the most exciting feature!
In some cases, the --regex-<LANG> and --mline-regex-<LANG> options are not
sufficient to generate the tags for a particular language. Some of the common
reasons for this are:
- To ignore commented lines or sections for the language file, so that tags aren’t generated for symbols that are within the comments.
- To enter and exit scope, and use it for tagging based on contextual state or with end-scope markers that are difficult to match to their associated scope entry point.
- To support nested scopes.
- To change the pattern searched for, or the resultant tag for the same pattern, based on scoping or contextual location.
- To break up an overly complicated
--mline-regex-<LANG>pattern into separate regex patterns, for performance or readability reasons.
To help handle such things, Universal Ctags has been enhanced with multi-table regex matching. The feature is inspired by lex, the fast lexical analyzer generator, which is a popular tool on Unix environments for writing parsers, and RegexLexer of Pygments. Knowledge about them will help you understand the new options.
The new options are:
--_tabledef-<LANG>- Declares a new regex matching table of a given name for the language, as described in “Declaring a new regex table”.
--_mtable-regex-<LANG>- Adds a regex pattern and associated tag generation information and flags, to the given table, as described in “Adding a regex to a regex table”.
--_mtable-extend-<LANG>- Includes a previously-defined regex table to the named one.
The above will be discussed in more detail shortly.
First, let’s explain the feature with an example. Consider an
imaginary language X has a similar syntax as JavaScript: var is
used as defining variable(s), and “/* ... */” is used for block
comments.
Here is our input, input.x:
/* BLOCK COMMENT
var dont_capture_me;
*/
var a /* ANOTHER BLOCK COMMENT */, b;
We want ctags to capture a and b - but it is difficult to write a parser
that will ignore dont_capture_me in the comment with a classical regex
parser defined with --regex-<LANG> or --mline-regex-<LANG>, because of
the block comments.
The --regex-<LANG> option only works on one line at a time, so can not know
dont_capture_me is within comments. The --mline-regex-<LANG> could
do it in theory, but due to the greedy nature of the regex engine it is
impractical and potentially inefficient to do so, given that there could be
multiple block comments in the file, with ‘*’ inside them, etc.
A parser written with multi-table regex, on the other hand, can capture only
a and b safely. But it is more complicated to understand.
Here is the 1st version of X.ctags:
Not so interesting. It doesn’t really do anything yet. It just creates a new
language named X, for files ending with a .x suffix, and defines a
new tag for variable kinds.
When writing a multi-table parser, you have to think about the necessary states of parsing. For the parser of language X, we need the following states:
- toplevel (initial state)
- comment (inside comment)
- vars (var statements)
Declaring a new regex table¶
Before adding regular expressions, you have to declare tables for each state
with the --_tabledef-<LANG>=<TABLE> option.
Here is the 2nd version of X.ctags doing so:
For table names, only characters in the range [0-9a-zA-Z_] are acceptable.
For a given language, for each file’s input the ctags multi-table parser begins
with the first declared table. For X.ctags, toplevel is the one.
The other tables are only ever entered/checked if another table specified to do
so, starting with the first table. In other words, if the first declared table
does not find a match for the current input, and does not specify to go to
another table, the other tables for that language won’t be used. The flags to go
to another table are {tenter}, {tleave}, and {tjump}, as described
later.
Adding a regex to a regex table¶
The new option to add a regex to a declared table is --_mtable-regex-<LANG>,
and it follows this form:
The parameters for --_mtable-regex-<LANG> look complicated. However,
<PATTERN>, <NAME>, and <KIND> are the same as the parameters of the
--regex-<LANG> and --mline-regex-<LANG> options. <TABLE> is simply
the name of a table previously declared with the --_tabledef-<LANG> option.
A regex pattern added to a parser with --_mtable-regex-<LANG> is matched
against the input at the current byte position, not line. Even if you do not
specify the ‘^’ anchor at the start of the pattern, ctags adds ‘^’ to
the pattern automatically. Unlike the --regex-<LANG> and
--mline-regex-<LANG> options, a ‘^’ anchor does not mean “beginning of
line” in --_mtable-regex-<LANG>; instead it means the beginning of the
input string (i.e., the current byte position).
The LONGFLAGS include the already discussed flags for --regex-<LANG> and
--mline-regex-<LANG>: {scope=...}, {mgroup=N}, {_advanceTo=N},
{basic}, {extend}, and {icase}. The {exclusive} flag does not
make sense for multi-table regex.
In addition, several new flags are introduced exclusively for multi-table regex use:
{tenter}- Push the current table on the stack, and enter another table.
{tleave}- Leave the current table, pop the stack, and go to the table that was just popped from the stack.
{tjump}- Jump to another table, without affecting the stack.
{treset}- Clear the stack, and go to another table.
{tquit}- Clear the stack, and stop processing the current input file for this language.
To explain the above new flags, we’ll continue using our example in the next section.
Skipping block comments¶
Let’s continue with our example. Here is the 3rd version of X.ctags:
Four --_mtable-regex-X lines are added for skipping the block comments. Let’s
discuss them one by one.
For each new file it scans, ctags always chooses the first pattern of the first table of the parser. Even if it’s an empty table, ctags will only try the first declared table. (in such a case it would immediately fail to match anything, and thus stop processing the input file and effectively do nothing)
The first declared table (toplevel) has the following regex added to
it first:
A pattern of \/\* is added to the toplevel table, to match the
beginning of a block comment. A backslash character is used in front of the
leading ‘/’ to escape the separation character ‘/’ that separates the fields
of --_mtable-regex-<LANG>. Another backslash inside the pattern is used
before the asterisk ‘*’, to make it a literal asterisk character in regex.
The last // means ctags should not tag something matching this pattern.
In --regex-<LANG> you never use // because it would be pointless to
match something and not tag it using and single-line --regex-<LANG>; in
multi-line --mline-regex-<LANG> you rarely see it, because it would rarely
be useful. But in multi-table regex it’s quite common, since you frequently
want to transition from one state to another (i.e., tenter or tjump
from one table to another).
The long flag added to our first regex of our first table is tenter, which
is a long flag for switching the table and pushing on the stack. {tenter=comment}
means “switch the table from toplevel to comment”.
So given the input file input.x shown earlier, ctags will begin at
the toplevel table and try to match the first regex. It will succeed, and
thus push on the stack and go to the comment table.
It will begin at the top of the comment table (it always begins at the top
of a given table), and try each regex line in sequence until it finds a match.
If it fails to find a match, it will pop the stack and go to the table that was
just popped from the stack, and begin trying to match at the top of that table.
If it continues failing to find a match, and ultimately reaches the end of the
stack, it will stop processing for this file. For the next input file, it will
begin again from the top of the first declared table.
Getting back to our example, the top of the comment table has this regex:
Similar to the previous toplevel table pattern, this one for \*\/ uses
a backslash to escape the separator ‘/’, as well as one before the ‘*’ to
make it a literal asterisk in regex. So what it’s looking for, from a simple
string perspective, is the sequence */. Note that this means even though
you see three backslashes /// at the end, the first one is escaped and used
for the pattern itself, and the --_mtable-regex-X only has // to
separate the regex pattern from the long flags, instead of the usual ///.
Thus it’s using the shorthand form of the --_mtable-regex-X option.
It could instead have been:
The above would have worked exactly the same.
Getting back to our example, remember we’re looking at the input.x
file, currently using the comment table, and trying to match the first
regex of that table, shown above, at the following location:
,ctags is trying to match starting here
v
/* BLOCK COMMENT
var dont_capture_me;
*/
var a /* ANOTHER BLOCK COMMENT */, b;
The pattern doesn’t match for the position just after /*, because that
position is a space character. So ctags tries the next pattern in the same
table:
This pattern matches any any one character including newline; the current
position moves one character forward. Now the character at the current position is
‘B’. The first pattern of the table */ still does not match with the input. So
ctags uses next pattern again. When the current position moves to the */
of the 3rd line of input.x, it will finally match this:
In this pattern, the long flag {tleave} is specified. This triggers table
switching again. {tleave} makes ctags switch the table back to the last
table used before doing {tenter}. In this case, toplevel is the table.
ctags manages a stack where references to tables are put. {tenter} pushes
the current table to the stack. {tleave} pops the table at the top of the
stack and chooses it.
So now ctags is back to the toplevel table, and tries the first regex
of that table, which was this:
It tries to match that against its current position, which is now the
newline on line 3, between the */ and the word var:
/* BLOCK COMMENT
var dont_capture_me;
*/ <--- ctags is now at this newline (/n) character
var a /* ANOTHER BLOCK COMMENT */, b;
The first regex of the toplevel table does not match a newline, so it tries
the second regex:
This matches a newline successfully, but has no actions to perform. So ctags
moves one character forward (the newline it just matched), and goes back to the
top of the toplevel table, and tries the first regex again. Eventually we’ll
reach the beginning of the second block comment, and do the same things as before.
When ctags finally reaches the end of the file (the position after b;),
it will not be able to match either the first or second regex of the
toplevel table, and quit processing the input file.
So far, we’ve successfully skipped over block comments for our new X
language, but haven’t generated any tags. The point of ctags is to generate
tags, not just keep your computer warm. So now let’s move onto actually tagging
variables…
Capturing variables in a sequence¶
Here is the 4th version of X.ctags:
One pattern in toplevel was added, and a new table vars with four
patterns was also added.
The new regex in toplevel is this:
The purpose of this being in toplevel is to switch to the vars table when
the keyword var is found in the input stream. We need to switch states
(i.e., tables) because we can’t simply capture the variables a and b
with a single regex pattern in the toplevel table, because there might be
block comments inside the var statement (as there are in our
input.x), and we also need to create two tags: one for a and one
for b, even though the word var only appears once. In other words, we
need to “remember” that we saw the keyword var, when we later encounter the
names a and b, so that we know to tag each of them; and saving that
“in-variable-statement” state is accomplished by switching tables to the
vars table.
The first regex in our new vars table is:
This pattern is used to match a single semi-colon ‘;’, and if it matches
pop back to the toplevel table using the {tleave} long flag. We
didn’t have to make this the first regex pattern, because it doesn’t overlap
with any of the other ones other than the /.// last one (which must be
last for this example to work).
The second regex in our vars table is:
We need this because block comments can be in variable definitions:
var a /* ANOTHER BLOCK COMMENT */, b;
So to skip block comments in such a position, the pattern \/\* is used just
like it was used in the toplevel table: to find the literal /* beginning
of the block comment and enter the comment table. Because we’re using
{tenter} and {tleave} to push/pop from a stack of tables, we can
use the same comment table for both toplevel and vars to go to,
because ctags will remember the previous table and {tleave} will
pop back to the right one.
The third regex in our vars table is:
This is nothing special, but is the one that actually tags something: it
captures the variable name and uses it for generating a variable (shorthand
v) tag kind.
The last regex in the vars table we’ve seen before:
This makes ctags ignore any other characters, such as whitespace or the
comma ‘,’.
Running our example¶
$ cat input.x
/* BLOCK COMMENT
var dont_capture_me;
*/
var a /* ANOTHER BLOCK COMMENT */, b;
$ u-ctags -o - --fields=+n --options=X.ctags input.x
u-ctags -o - --fields=+n --options=X.ctags input.x
a input.x /^var a \/* ANOTHER BLOCK COMMENT *\/, b;$/;" v line:4
b input.x /^var a \/* ANOTHER BLOCK COMMENT *\/, b;$/;" v line:4
It works!
You can find additional examples of multi-table regex in our github repo, under
the optlib directory. For example puppetManifest.ctags is a serious
example. It is the primary parser for testing multi-table regex parsers, and
used in the actual ctags program for parsing puppet manifest files.
Scheduling a guest parser with _guest regex flag¶
With _guest regex flag, you can run a parser (a guest parser) on an
area of the current input file.
See “Guest parser: Applying a parser to specified areas of input file” about the concept of the guest parser.
The _guest regex flag specifies a guest spec, and attaches it to
the associated regex pattern.
A guest spec has three fields: <PARSER>, <START> of area, and <END> of area.
The _guest regex flag has following forms:
{_guest=<PARSER>,<START>,<END>}
ctags maintains a data called guest request during parsing. A guest request also has three fields: parser, start of area, and end of area.
You, a parser developer, have to fill the fields of guest specs. ctags inquiries the guest spec when matching the regex pattern associated with it, tries to fill the fields of the guest request, and runs a guest parser when all the fields of the guest request are filled.
If you use Multi-line pattern match to define a host parser, you must specify all the fields of guest request.
On the other hand if you don’t use Multi-line pattern match to define a host parser, ctags can fill fields of guest request incrementally; more than one guest specs are used to fill the fields. In other words, you can make some of the fields of a guest spec empty.
The <PARSER> field of _guest regex flag¶
For <PARSER>, you can specify one of the following items:
a name of a parser
If you know the guest parser you want to run before parsing the input file, specify the name of the parser.
An example of running C parser as a guest parser:
{_guest=C,...
the group number of a regex pattern started from ‘\’ (backslash)
If a parser name appears in an input file, write a regex pattern to capture the name. Specify the group number where the name is stored to the parser. In such case, use ‘
\’ as the prefix for the number.Let’s see an example. Git Flavor Markdown (GFM) is a language for documentation. It provides a notation for quoting a snippet of program code; the language treats the area started from
~~~to~~~as a snippet. You can specify a programming language of the snippet with starting the area with~~~<THE_NAME_OF_LANGUAGE>, like~~~Cor~~~Java.To run a guest parser on the area, you have to capture the <THE_NAME_OF_LANGUAGE> with a regex pattern:
The pattern captures the language name in the input file with the regex group 1, and specify it to <PARSER>:
{guest=\1,...
the group number of a regex pattern started from ‘*’ (asterisk)
If a file name implying a programming language appears in an input file, capture the file name with the regex pattern where the guest spec attaches to. ctags tries to find a proper parser for the file name by inquiring the langmap.
Use ‘
*’ as the prefix to the number for specifying the group of the regex pattern that captures the file name.Let’s see an example. Consider you have a shell script that emits a program code instantiated from one of the templates. Here documents are used to represent the templates like:
i=... cat > foo.c <<EOF int main (void) { return $i; } EOF cat > foo.el <<EOF (defun foo () (1+ $i)) EOFTo run guest parsers for the here document areas, the shell script parser of ctags must choose the parsers from the file names (
foo.candfoo.el):The pattern captures the file name in the input file with the regex group 1, and specify it to <PARSER>:
{_guest=*1,...
The <START> and <END> fields of _guest regex flag¶
The <START> and <END> fields specify the area the <PARSER> parses. <START> specifies the start of the area. <END> specifies the end of the area.
The forms of the two fields are the same: a regex group number
followed by start or end. e.g. 3start, 0end. The suffixes,
start and end, represents one of two boundaries of the group.
Let’s see an example:
{_guest=C,2end,3start}
This guest regex flag means running C parser on the area between
2end and 3start. 2end means the area starts from the end of
matching of the 2nd regex group associated with the flag. 3start
means the area ends at the beginning of matching of the 3rd regex
group associated with the flag.
Let’s more realistic example. Here is an optlib file for an imaginary language single:
This parser can run C parser and extract main function from the
following input file:
BEGIN_C<int main (int argc, char **argv) { return 0; }>END_C
^ ^
`- "1end" points here. |
"2start" points here. -+
Defining a subparser¶
Basic¶
About the concept of subparser, see “Subparser: Tagging definitions of higher (upper) level language”.
--langdef=<LANG> option is extended as
--langdef=<LANG>[{base=<LANG>}[{shared|dedicated|bidirectional}]][{_autoFQTag}] to define
a subparser for a specified base parser. Combining with --kinddef-<LANG>
and --regex-<KIND> options, you can extend an existing parser
without risk of kind confliction.
Let’s see an example.
input.c
static int set_one_prio(struct task_struct *p, int niceval, int error)
{
}
SYSCALL_DEFINE3(setpriority, int, which, int, who, int, niceval)
{
...;
}
$ ctags -x --_xformat="%20N %10K %10l" -o - input.c
set_one_prio function C
SYSCALL_DEFINE3 function C
C parser doesn’t understand that SYSCALL_DEFINE3 is a macro for defining an
entry point for a system.
Let’s define linux subparser which using C parser as a base parser (linux.ctags):
The output is change as follows with linux parser:
$ ctags --options=./linux.ctags -x --_xformat="%20N %10K %10l" -o - input.c
setpriority syscall linux
set_one_prio function C
SYSCALL_DEFINE3 function C
setpriority is recognized as a syscall of linux.
Using only --regex-C=... you can capture setpriority.
However, there were concerns about kind confliction; when introducing
a new kind with --regex-C=..., you cannot use a letter and name already
used in C parser and --regex-C=... options specified in the other places.
You can use a newly defined subparser as a new namespace of kinds.
In addition you can enable/disable with the subparser usable
--languages=[+|-] option:
Direction flags¶
As explained in “Direction flags” in “Running multiple parsers on an input file”, you can choose direction(s) how a base parser and a guest parser work together with direction flags.
The following examples are taken from #1409 submitted by @sgraham on github Universal Ctags repository.
input.cc and input.mojom are input files, and have the same
contents:
ABC();
int main(void)
{
}
C++ parser can capture main as a function. Mojom subparser defined in the
later runs on C++ parser and is for capturing ABC.
dedicated combination¶
{dedicated} is specified, for input.cc, only tags capture by C++
parser are recorded to tags file. For input.mojom, both tags capture
by C++ parser and mojom parser are recorded to tags file.
mojom-dedicated.ctags:
Mojom parser works only when .mojom file is given as input.
bidirectional combination¶
{bidirectional} is specified, both tags capture by C++ parser and
mojom parser are recorded to tags file for either input input.cc and
input.mojom.
mojom-bidirectional.ctags:
Translating an option file into C source code (optlib2c)¶
Universal Ctags has an optlib2c script that translates an option file into C
source code. Your optlib parser can thus easily become a built-in parser.
To add your optlib file, foo.ctags, into ctags do the following steps;
- copy
foo.ctagsfile onoptlib/directory - add
foo.ctagsonOPTLIB2C_INPUTvariable insource.mak - add
fooParseronPARSER_LISTmacro variable inmain/parser_p.h
You are encouraged to submit your .ctags file to our repository on
github through a pull request. See “Contributions” for more details.