|
|
The macro language is a subset of m4. The description of the languagecomes from the m4 man page, with annotations in square brackets.
Macro calls have the form:
name(arg1,arg2, ..., argn)
The left parenthesis (() must immediately follow the name of the macro. If a defined macro name is not followed by a left parenthesis ((), it is deemed to have no arguments. Leading unquoted blanks, tabs, and newlines are ignored while collecting arguments. Potential macro names consist of alphabetic letters, digits, and underscore _, where the first character is not a digit.
[ Suffixing an empty set of parentheses to a macro, e.g., `foo()' results in a macro called with one argument which happens to be null. This is not the same as `foo', which is a macro called with zero arguments, although the only way to tell the difference is to look at `$#' or `$@'. ]
Left and right single quotation marks are used to quote strings. The value of a quoted string is the string stripped of the quotation marks.
[ Quotation marks *do* nest. ]
When a macro name is recognized, its arguments are collected by searching for a matching right parenthesis.
[ `Matching' means that nested parentheses are respected during argument scanning. ]
Macro evaluation proceeds normally during the collection of the arguments, and any commas or right parentheses which happen to turn up within the value of a nested call are as effective as those in the original input text. After argument collection, the value of the macro is pushed back onto the input stream and rescanned.
[ This means that all macro arguments are scanned at least once. There is no implicit quoting. This means that you probably want to quote the arguments to most of the builtin macros. For example, assuming that `foo' and `baz' are not already defined, then the input
define(foo, bar) define(baz, blurfl) define(foo, baz) undefine(foo, baz)
first defines macros `foo' (which expands to `bar') and `baz' (which expands to `blurfl'). The third line defines a macro named `baz' which expands to `blurfl', because the arguments are macro-expanded before interpretation. And similarly, the fourth line removes the definitions of `bar' and `blurfl'. Play it safe. Quote everything that should not be expanded. Adding quotation marks
define(`foo', `bar') define(`baz', `blurfl') define(`foo', `baz') undefine(`foo', baz')
does not change the meanings of the first two lines, but the second line redefines `foo' to expand to `baz', and the third line removes the definitions of `foo' and `baz'.
Note also that the value of the macro is inspected *after* argument expansion is complete. This means that
define(`foo', `bar') foo(define(`foo', `baz'))
expands to `baz'. However, under the AT&T implementation,
define(`foo', `bar') foo(undefine(`foo')) still expands to `bar' for some reason I do not understand.
Even worse, the GNU implementation displays garbage!
My implementation expands this to `foo'.
Similarly,
define(`foo', `bar') pushdef(`foo', `baz') foo(popdef(`foo'))
expands to `baz' under AT&T and to garbage under GNU. I expand this to `bar'. ]
M4 makes available the following built-in macros. They may be redefined, but once this is done the original meaning is lost. Their values are null unless otherwise stated:
define The second argument is installed as the value of the macro whose name is the first argument. Each occurrence of $n in the replacement text, where n is a digit, is replaced by the n-th argument. Argument 0 is the name of the macro; missing arguments are replaced by the null string; $# is replaced by the number of arguments; $* is replaced by a list of all the arguments separated by commas; @ $@ is like $*, but each argument is quoted (with the current quotation marks).
[ Don't forget to quote both arguments unless you really know what you're doing. Note also that macro substitution does not respect quotation marks. Therefore, the macro
define(`quote',``$1'')
quotes its argument. Arguments beyond the second are ignored. If passed no arguments, does nothing. ]
undefine Removes the definition of the macro named in its argument.
[ Not documented is that you can pass multiple names to `undefine' and they will all become undefined. Again, remember to quote the arguments.
Undefine'ing a macro also destroys its popdef stack. ]
defn Returns the quoted definition of its argument(s). It is useful for renaming macros, especially built-ins.
[ Also handy for augmenting an existing macro. Again, remember to quote the arguments.
Implementation note: AT&T m4 uses bytes with the high bit set to represent built-ins. If an input file contains characters with the high bit set, AT&T m4 may core dump.
This implementation uses byte 0x00 to represent built-ins. An occurrence of 0x00 in the input stream is quoted to 0x00 0x00 so that the out-of-band marker is truly out-of-band.
[ SOMEDAY - Except that they don't look pretty in the output ]
AT&T and GNU do not support embedding defn's of builtins into macros; you must do a pure definition or it doesn't count. For example, you can't say
define(`defx',defn(`define')`(x,`$1')')
to make `defx' define the symbol `x' independent of what happened to the macro `define'.
I support this.
]
pushdef Like define, but saves any previous definition.
popdef Removes current definition of its argument(s), exposing the previous one if any.
[ If no definition was pushed, then it acts like `undefine'. ]
ifdef If the first argument is defined, the value is the second argument, otherwise the third. If there is no third argument, the value is null.
[ Again, remember to quote the first argument. And you probably want to quote the second and third arguments to delay their evaluation until time they are actually needed. ]
shift Returns all but its first argument. The other arguments are quoted and pushed back with commas in between. The quoting nullifies the effect of the extra scan that will subsequently be performed.
[ Shift is typically used as `shift($@)' to delete the first argument. ]
changequote Changes quotation marks to the first and second arguments. The symbols may be up to five characters long. Changequote without arguments restores the original values.
[ Not implemented. Quotation marks may not contain alphanumerics, underscores, or whitespace. ]
changecom Changes left and right comment markers from the default # and newline. With no arguments, the comment mechanism is effectively disabled. With one argument, the left marker becomes the argument and the right marker becomes newline. With two arguments, both markers are affected. Comment markers may be up to five characters long.
[ Not implemented. Comment marks may not contain alphanumerics, underscores, or whitespace. ]
divert M4 maintains 10 output streams, numbered 0-9. The final output is the concatenation of the streams in numerical order; initially stream 0 is the current stream. The divert macro changes the current output stream to its (digit-string) argument. Output diverted to a stream other than 0 through 9 is discarded.
[ Only diversions 0 and -1 are currently supported. ]
undivert Causes immediate output of text from diversions named as arguments, or all diversions if no argument. Text may be undiverted into another diversion. Undiverting discards the diverted text.
[ Not implemented.
Text being undiverted is not subject to macro scanning. ]
divnum Returns the value of the current output stream.
[ If output is being discarded, then divnum returns -1. ]
dnl Reads and discards characters up to and including the next newline.
[ dnl is used to embed m4 comments into the file. ]
ifelse Has three or more arguments. If the first argument is the same string as the second, then the value is the third argument. If not, and if there are more than four arguments, the process is repeated with arguments 4, 5, 6 and 7. Otherwise, the value is either the fourth string, or if it is not present, null.
[ In other words,
ifelse(l1,r1,v1,l2,r2,v2,...,ln,rn,vn,v)
If l1 equals r1, then return v1. Else if l2 equals r2, then return v2. ... Else if ln equals rn, then return vn. Else return v (if provided).
Note that all the v's will be expanded at least once (during scanning of arguments), and the winning v will be expanded twice (the second during rescan). Therefore, you probably want to quote all the v's. Similarly, all the l's and r's will be expanded exactly once, even the ones that are never used.
AT&T does not raise an error if fewer than three arguments are provided. ]
incr Returns the value of its argument incremented by 1. The value of the argument is calculated by interpreting an initial digit-string as a decimal number.
decr Returns the value of its argument decremented by 1.
eval Evaluates its argument as an arithmetic expression, using 32-bit arithmetic. Operators include +, -, *, /, %, ^ (exponentiation), bitwise &, |, ^, and ~; relationals; parentheses. Octal and hex numbers may be specified as in C. The second argument specifies the radix for the result; the default is 10. The third argument may be used to specify the minimum number of digits in the result.
[ The documentation on eval operators is wrong. Eval supports C operators, which means that ^ denotes bitwise exclusive-or rather than exponentiation. The exponentiation operator is `**'.
AT&T does not support the ternary ?: operator.
A radix less than 2 is treated as radix 10.
AT&T considers `0x' to eval to zero. I consider it an error.
AT&T allows 8 and 9 as octal digits. I don't.
AT&T generates peculiar output when given extremely large radices, sometimes resulting in a core dump. Try `eval(100,101)' for example.
This implementation will treat radices greater than 201 as errors. 201 is the magic number, because base 201 uses digits 0' ... '9', then 'A' ... '\377'.
Since a leading zero causes the value to be parsed as octal, this causes problems when parsing values which have been zero-padded. To force evaluation as decimal, you can use `incr($1)-1', since `incr' accepts only decimal input. ]
len Returns the number of characters in its argument.
[ AT&T does not raise an error if too many or too few arguments are passed. ]
index Returns the position in its first argument where the second argument begins (zero origin), or -1 if the second argument does not occur.
[ AT&T does not raise an error if too many or too few arguments are passed. If `index' appears as a bareword, AT&T generates a null expansion. GNU emits the word `index'.
I side with GNU on this, only because the word `index' frequently appears in plaintext and it would be bad to see it disappear randomly. (On the grounds of purity, I should raise an error.) ]
substr Returns a substring of its first argument. The second argument is a zero origin number selecting the first character; the third argument indicates the length of the substring. A missing third argument is taken to be large enough to extend to the end of the first string.
translit Transliterates the characters in its first argument from the set given by the second argument to the set given by the third. No abbreviations are permitted.
[ GNU supports ranges in the translit, e.g., 0-9 as shorthand for 0123456789. I do not.
As a convenience, a bare `translit' is left unexpanded. (GNU)
]
include Returns the contents of the file named in the argument.
[ It is not an error if the argument is the null string, in which case no file is included.
As a convenience, a bare `include' is left unexpanded. (GNU)
]
sinclude Identical to include, except that it says nothing if the file is inaccessible.
syscmd Executes the UNIX command given in the first argument. No value is returned.
[ Not implemented. ]
sysval Is the return code from the last call to syscmd.
[ Not implemented. ]
maketemp Fills in a string of XXXXX in its argument with the current process ID.
[ Not implemented. ]
m4exit Causes immediate exit from m4. Argument 1, if given, is the exit code; the default is 0.
[ Not implemented. ]
m4wrap Argument 1 will be pushed back at final end-of- file; for example, m4wrap(`cleanup()').
[ Not implemented.
The AT&T implementation supports only one wrapup token. If you try to wrap twice, the second overwrites the first. ]
errprint Prints its argument on the diagnostic output file.
dumpdef Prints current names and definitions, for the named items, or for all if no arguments are given.
[ The dump is made to the diagnostic output file. The format of the output is
macroname:\tvalue
where \t is an ASCII `tab' character. Built-ins are dumped as
<originalname>
so, for example, after `define(`foo',`baz'defn(`define')`blatz')' a `dumpdef(`foo')' will display
foo: baz<define>blatz
]
traceon With no arguments, turns on tracing for all macros (including built-ins). Otherwise, turns on tracing for named macros.
[ Not fully implemented.
Trace output is of the form
Trace(n): macroname(args)
where `n' is the macro depth (top-level is depth 0).
We currently do not dump the depth.
]
traceoff Turns off trace globally and for any macros specified. Macros specifically traced by traceon can be untraced only by specific calls to traceoff.
[ In other words, each macro has a specific `trace' bit which is controlled by specifying the macro's name in the argument list of a `traceon' or `traceoff'. There is also a global trace bit which is turned on by `traceon' with no arguments, and is turned off by `traceoff' with any number of arguments.
A macro is traced if the global trace bit is on, or if the macro's private trace bit is on.
The state of the macro's trace bit is pushed and popped by `pushdef' and `popdef'. If a macro is defined, pushdef'd or undefined, its trace bit is reset. ]
[ patsubst is a GNU extension which is partially implemented. (Just barely enough to keep the build happy.)
patsubst Starting with $1, replace all occurrences of $2 with $3. If $3 is omitted, it is taken as the null string. If $2 is the null string, then $3 is inserted before each character of $1.
NOTE! That this is not the same as GNU. In GNU, $2 is a regular expression. In our implementation, it is merely a literal. ]
******************************************************************************
Implementation details
Input streams
A stack of input streams is maintained, up to the file handle limit imposed by the operating system. The `include' and `sinclude' macros push a new file onto the head of the stream. Macro expansion pushes a string onto the head of the stream.
Comments and quoted strings
Comments and quoted strings are essentially the same thing: They enclose text that should be passed through without interpretation. The difference is that quotation marks nest and are stripped, whereas comments do not nest and are not stripped.
|
