Merge branch 'cooking' of github.com:graphitemaster/gmqcc into cooking

[xonotic/gmqcc.git] / doc / gmqcc.1

.\"mdoc
.Dd January 24, 2013
.Dt GMQCC 1 PRM
.Os
.Sh NAME
.Nm gmqcc
.Nd A Quake C compiler built from the NIH realm of sarcastic wit
.Sh SYNOPSIS
.Nm gmqcc
.Op Cm options
.Op Ar files...
.Sh DESCRIPTION
Traditionally, a QC compiler reads the file
.Pa progs.src
which in its first line contains the output filename, and the rest is a
list of QC source files that are to be compiled in order.
.Nm gmqcc
optionally takes options to specify the output and
input files on the commandline, and also accepts assembly files.
.Sh OPTIONS
.Nm gmqcc
mostly tries to mimic gcc's commandline handling, though
there are also traditional long-options available.
.Bl -tag -width Ds
.It Fl h , Fl -help
Show a usage message and exit.
.It Fl o , Fl -output= Ns Ar filename
Specify the output filename. Defaults to progs.dat. This will overwrite
the output file listed in a
.Pa progs.src
file in case such a file is used.
.Bl -tag -width indent
.It Fl O Ns Ar number
Specify the optimization level
.It Ar 3
Highest optimization level
.It Ar 2
Default optimization level
.It Ar 1
Minimal optimization level
.It Ar 0
Disable optimization entirely
.El
.Pp
.It Fl O Ns Ar name , Fl Ono- Ns Ar name
Enable or disable a specific optimization. Note that these options
must be used after setting the optimization level, otherwise they'll
be overwritten.
.It Fl O Ns Cm help
List all possible optimizations and the optimization level they're
activated at.
.It Fl q , Fl -quiet
Be less verbose. In particular removes the messages about which files
are being processed, and which compilation mode is being used, and
some others. Warnings and errors will of course still be displayed.
.It Fl D Ns Ar macroname , Fl D Ns Ar macroname Ns = Ns Ar value
Predefine a macro, optionally with a optional value.
.It Fl E
Run only the preprocessor as if
.Fl f Ns Cm ftepp
was used and print the preprocessed code to stdout.
.It Fl W Ns Ar warning , Fl Wno- Ns Ar warning
Enable or disable a warning.
.It Fl W Ns Cm all
Enable almost all warnings. Overrides preceding
.Fl W
parameters.
.Pp
The following warnings will
.Em not
be enabled:
.Bl -tag -width indent -offset indent
.It Fl W Ns Cm uninitialized-global
.El
.It Fl W Ns Cm error , Fl Wno- Ns Cm error
Controls whether or not all warnings should be treated as errors.
.It Fl Werror- Ns Ar warning , Fl Wno-error- Ns Ar warning
Controls whether a specific warning should be an error.
.It Fl W Ns Cm help
List all possible warn flags.
.It Fl f Ns Ar flag , Fl fno- Ns Ar flag
Enable or disable a specific compile flag. See the list of flags
below.
.It Fl f Ns Cm help
List all possible compile flags.
.It Fl nocolor
Disables colored output
.It Fl config= Ns Ar file
Use an ini file to read all the
.Fl O , Fl W
and
.Fl f
flag from. See the
.It Fl "debug"
Turn on some compiler debugging mechanisms.
.It Fl memchk
Turn on compiler mem-check. (Shows allocations and checks for leaks.)
.It Fl -memdumpcols Ns Ar columns
Changes the number of columns to use for the debug memory dump, defaults to 16.
.Sx CONFIG
section about the file format.
.It Fl redirout= Ns Ar file
Redirects standard output to a
.Ar file
.It Fl redirerr= Ns Ar file
Redirects standard error to a
.Ar file
.It Fl std= Ns Ar standard
Use the specified standard for parsing QC code. The following standards
are available:
.Ar gmqcc , Ar qcc , Ar fteqcc
Selecting a standard also implies some
.Fl f
options and behaves as if
those options have been written right after the
.Fl std
option, meaning
if you changed them before the
.Fl -std
option, you're now overwriting them.
.Pp
.Fl std= Ns Cm gmqcc No includes:
.Bl -tag -width indent -compact -offset Ds
.It Fl f Ns Cm adjust-vector-fields
.It Fl f Ns Cm correct-logic
.It Fl f Ns Cm true-empty-strings
.It Fl f Ns Cm loop-labels
.It Fl f Ns Cm initialized-nonconstants
.It Fl f Ns Cm translatable-strings
.It Fl fno- Ns Cm false-empty-strings
.It Fl W Ns Cm invalid-parameter-count
.It Fl W Ns Cm missing-returnvalues
.It Fl f Ns Cm correct-ternary Li (cannot be turned off)
.El
.Pp
.Fl std= Ns Cm qcc No includes:
.Bl -tag -width indent -compact -offset Ds
.It Fl f Ns Cm assign-function-types
.It Fl fIno- Ns Cm adjust-vector-fields
.El
.Pp
.Fl std= Ns Cm fteqcc No includes:
.Bl -tag -width indent -compact -offset Ds
.It Fl f Ns Cm ftepp
.It Fl f Ns Cm translatable-strings
.It Fl f Ns Cm assign-function-types
.It Fl W Ns Cm ternary-precedence
.It Fl fno- Ns Cm adjust-vector-fields
.It Fl fno- Ns Cm correct-ternary
.El
.It Fl -add-info
Adds compiler information to the generated binary file. Currently
this includes the following globals:
.Bl -tag -width indent -compact
.It Li reserved:version
String containing the compiler version as printed by the --version
parameter.
.El
.It Fl -correct , Fl -no-correct
When enabled, errors about undefined values try to suggest an existing
value via spell checking.
.It Fl dump
DEBUG OPTION. Print the code's intermediate representation before the
optimization and finalization passes to stdout before generating the
binary.
.It Fl dumpfin
DEBUG OPTION. Print the code's intermediate representation after the
optimization and finalization passes to stdout before generating the
binary. The instructions will be enumerated, and values will contain a
list of liferanges.
.El
.Sh COMPILE WARNINGS
.Bl -tag -width Ds
.It Fl W Ns Cm unused-variable
Generate a warning about variables which are declared but never used.
This can be avoided by adding the
.Ql noref
keyword in front of the
variable declaration. Additionally a complete section of unreferenced
variables can be opened using
.Ql #pragma noref 1
and closed via
.Ql #pragma noref 0 Ns .
.It Fl W Ns Cm used-uninitialized
Generate a warning if it is possible that a variable can be used
without prior initialization. Note that this warning is not
necessarily reliable if the initialization happens only under certain
conditions. The other way is
.Em not
possible: that the warning is
.Em not
generated when uninitialized use
.Em is
possible.
.It Fl W Ns Cm unknown-control-sequence
Generate an error when an unrecognized control sequence in a string is
used. Meaning: when there's a character after a backslash in a string
which has no known meaning.
.It Fl W Ns Cm extensions
Warn when using special extensions which are not part of the selected
standard.
.It Fl W Ns Cm field-redeclared
Generally QC compilers ignore redeclaration of fields. Here you can
optionally enable a warning.
.It Fl W Ns Cm missing-return-values
Functions which aren't of type
.Ft void
will warn if it possible to
reach the end without returning an actual value.
.It Fl W Ns Cm invalid-parameter-count
Warn about a function call with an invalid number of parameters.
.It Fl W Ns Cm local-shadows
Warn when a locally declared variable shadows variable.
.It Fl W Ns Cm local-constants
Warn when the initialization of a local variable turns the variable
into a constant. This is default behaviour unless
.Fl f Ns Cm initialized-nonconstants
is used.
.It Fl W Ns Cm void-variables
There are only 2 known global variables of type void:
.Ql end_sys_globals
and
.Ql end_sys_fields Ns .
Any other void-variable will warn.
.It Fl W Ns Cm implicit-function-pointer
A global function which is not declared with the
.Ql var
keyword is
expected to have an implementing body, or be a builtin. If neither is
the case, it implicitly becomes a function pointer, and a warning is
generated.
.It Fl W Ns Cm variadic-function
Currently there's no way for an in QC implemented function to access
variadic parameters. If a function with variadic parameters has an
implementing body, a warning will be generated.
.It Fl W Ns Cm frame-macros
Generate warnings about
.Ql $frame
commands, for instance about
duplicate frame definitions.
.It Fl W Ns Cm effectless-statement
Warn about statements which have no effect. Any expression which does
not call a function or assigns a variable.
.It Fl W Ns Cm end-sys-fields
The
.Ql end_sys_fields
variable is supposed to be a global variable
of type
.Ft void Ns .
It is also recognized as a \fIfield\fR but this
will generate a warning.
.It Fl W Ns Cm assign-function-types
Warn when assigning to a function pointer with an unmatching
signature. This usually happens in cases like assigning the null
function to an entity's .think function pointer.
.It Fl W Ns Cm cpp
Show warnings created using the preprocessor's '#warning' directive.
.It Fl W Ns Cm multifile-if
Warn if there's a preprocessor \fI#if\fR spanning across several
files.
.It Fl W Ns Cm double-declaration
Warn about multiple declarations of globals. This seems pretty common
in QC code so you probably do not want this unless you want to clean
up your code.
.It Fl W Ns Cm const-var
The combination of \fIconst\fR and \fIvar\fR is not illegal, however
different compilers may handle them differently. We were told, the
intention is to create a function-pointer which is not assignable.
This is exactly how we interpret it. However for this interpretation
the
.Ql var
keyword is considered superfluous (and philosophically
wrong), so it is possible to generate a warning about this.
.It Fl W Ns Cm multibyte-character
Warn about multibyte character constants, they do not work right now.
.It Fl W Ns Cm ternary-precedence
Warn if a ternary expression which contains a comma operator is used
without enclosing parenthesis, since this is most likely not what you
actually want. We recommend the
.Fl f Ns Cm correct-ternary
option.
.It Fl W Ns Cm unknown-pragmas
Warn when encountering an unrecognized
.Ql #pragma
line.
.It Fl W Ns Cm unreachable-code
Warn about unreachable code. That is: code after a return statement,
or code after a call to a function marked as 'noreturn'.
.It Fl W Ns Cm debug
Enable some warnings added in order to help debugging in the compiler.
You won't need this.
.It Fl W Ns Cm unknown-attribute
Warn on an unknown attribute. The warning will inlclude only the first
token inside the enclosing attribute-brackets. This may change when
the actual attribute syntax is better defined.
.It Fl W Ns Cm reserved-names
Warn when using reserved names such as
.Ql nil Ns .
.It Fl W Ns Cm uninitialized-constant
Warn about global constants (using the
.Ql const
keyword) with no
assigned value.
.It Fl W Ns Cm uninitialized-global
Warn about global variables with no initializing value. This is off by
default, and is added mostly to help find null-values which are
supposed to be replaced by the untyped 'nil' constant.
.It Fl W Ns Cm different-qualifiers
Warn when a variables is redeclared with a different qualifier. For
example when redeclaring a variable as \'var\' which was previously
marked \'const\'.
.It Fl W Ns Cm different-attributes
Similar to the above but for attributes like
.Ql [[noreturn]] Ns .
.It Fl W Ns Cm deprecated
Warn when a function is marked with the attribute
"[[deprecated]]". This flag enables a warning on calls to functions
marked as such.
.It Fl W Ns Cm parenthesis
Warn about possible mistakes caused by missing or wrong parenthesis,
like an assignment in an 'if' condition when there's no additional set
of parens around the assignment.
.It Fl W Ns Cm unsafe-types
When passing variadic parameters via
.Li ...(N)
it can happen that incompatible types are passed to functions. This
enables several warnings when static typechecking cannot guarantee
consistent behavior.
.It Fl W Ns Cm breakdef
When compiling original id1 QC there is a definition for `break`
which conflicts with the 'break' keyword in GMQCC. Enabling this
will print a warning when the definition occurs. The definition is
ignored for both cases.
.It Fl W Ns Cm const-overwrite
When compiling original QuakeWorld QC there are instances where
code overwrites constants. This is considered an error, however
for QuakeWorld to compile it needs to be treated as a warning
instead, as such this warning only works when -std=qcc.
.El
.Sh COMPILE FLAGS
.Bl -tag -width Ds
.It Fl f Ns Cm darkplaces-string-table-bug
Add some additional characters to the string table in order to
compensate for a wrong boundcheck in some specific version of the
darkplaces engine.
.It Fl f Ns Cm adjust-vector-fields
When assigning to field pointers of type \fI.vector\fR the common
behaviour in compilers like \fIfteqcc\fR is to only assign the
x-component of the pointer. This means that you can use the vector as
such, but you cannot use its y and z components directly. This flag
fixes this behaviour. Before using it make sure your code does not
depend on the buggy behaviour.
.It Fl f Ns Cm ftepp
Enable a partially fteqcc-compatible preprocessor. It supports all the
features used in the Xonotic codebase. If you need more, write a
ticket.
.It Fl f Ns Cm ftepp-predefs
Enable some predefined macros. This only works in combination with
\'-fftepp' and is currently not included by '-std=fteqcc'. The
following macros will be added:
.Bd -literal -offset indent
__LINE__
__FILE__
__COUNTER__
__COUNTER_LAST__
__RANDOM__
__RANDOM_LAST__
__DATE__
__TIME__
__FUNC__
.Ed
.Pp
Note that
.Li __FUNC__
is not actually a preprocessor macro, but is recognized by the parser
even with the preprocessor disabled.
.Pp
Note that fteqcc also defines
.Li __NULL__
which becomes the first global. Assigning it to a vector does not
yield the same result as in gmqcc where
.Li __NULL__
is defined to
.Li nil
(See
.Fl f Ns Cm untyped-nil
), which will cause the vector to be zero in all components. With fteqcc
only the first component will be 0, while the other two will become
the first to of the global return value. This behavior is odd and
relying on it should be discouraged, and thus is not supported by
gmqcc.
.It Fl f Ns Cm relaxed-switch
Allow switch cases to use non constant variables.
.It Fl f Ns Cm short-logic
Perform early out in logical AND and OR expressions. The final result
will be either a 0 or a 1, see the next flag for more possibilities.
.It Fl f Ns Cm perl-logic
In many languages, logical expressions perform early out in a special
way: If the left operand of an AND yeilds true, or the one of an OR
yields false, the complete expression evaluates to the right side.
Thus
.Ql true && 5
evaluates to 5 rather than 1.
.It Fl f Ns Cm translatable-strings
Enable the underscore intrinsic: Using
.Ql _("A string constant")
will cause the string immediate to get a name with a "dotranslate_"
prefix. The darkplaces engine recognizes these and translates them in
a way similar to how gettext works.
.It Fl f Ns Cm initialized-nonconstants
Don't implicitly convert initialized variables to constants. With this
flag, the \fIconst\fR keyword is required to make a constant.
.It Fl f Ns Cm assign-function-types
If this flag is not set, (and it is set by default in the qcc and
fteqcc standards), assigning function pointers of mismatching
signatures will result in an error rather than a warning.
.It Fl f Ns Cm lno
Produce a linenumber file along with the output .dat file.
.It Fl f Ns Cm correct-ternary
Use C's operator precedence for ternary expressions. Unless your code
depends on fteqcc-compatible behaviour, you'll want to use thi
soption.
.It Fl f Ns Cm single-vector-defs
Normally vectors generate 4 defs, once for the vector, and once for
its components with _x, _y, _z suffixes. This option
prevents components from being listed.
.It Fl f Ns Cm correct-logic
Most QC compilers translate
.Ql if(a_vector)
directly as an IF on the
vector, which means only the x-component is checked. This option causes
vectors to be cast to actual booleans via a NOT_V and, if necessary, a
NOT_F chained to it.
.Bd -literal -offset indent
if (a_vector) // becomes
if not(!a_vector)
// likewise
a = a_vector && a_float // becomes
a = !!a_vector && a_float
.Ed
.It Fl f Ns Cm true-empty-strings
An empty string is considered to be true everywhere. The NOT_S
instruction usually considers an empty string to be false, this option
effectively causes the unary not in strings to use NOT_F instead.
.It Fl f Ns Cm false-empty-strings
An empty string is considered to be false everywhere. This means loops
and if statements which depend on a string will perform a NOT_S
instruction on the string before using it.
.It Fl f Ns Cm utf8
Enable utf8 characters. This allows utf-8 encoded character constants,
and escape sequence codepoints in the valid utf-8 range. Effectively
enabling escape sequences like '\\{x2211}'.
.It Fl f Ns Cm bail-on-werror
When a warning is treated as an error, and this option is set (which
it is by default), it is like any other error and will cause
compilation to stop. When disabling this flag by using
\-fno-bail-on-werror, compilation will continue until the end, but no
output is generated. Instead the first such error message's context is
shown.
.It Fl f Ns Cm loop-labels
Allow loops to be labeled, and allow 'break' and 'continue' to take an
optional label to decide which loop to actually jump out of or
continue.
.Bd -literal -offset indent
for :outer (i = 0; i < n; ++i) {
    while (inner) {
        ...;
        if (something)
            continue outer;
    }
}
.Ed
.It Fl f Ns Cm untyped-nil
Adds a global named 'nil' which is of no type and can be assigned to
anything. No typechecking will be performed on assignments. Assigning
to it is forbidden, using it in any other kind of expression is also
not allowed.
.sp
Note that this is different from fteqcc's __NULL__: In fteqcc,
__NULL__ maps to the integer written as '0i'. It's can be assigned to
function pointers and integers, but it'll error about invalid
instructions when assigning it to floats without enabling the FTE
instruction set. There's also a bug which allows it to be assigned to
vectors, for which the source will be the global at offset 0, meaning
the vector's y and z components will contain the OFS_RETURN x and y
components.
.sp
In that gmqcc the nil global is an actual global filled with zeroes,
and can be assigned to anything including fields, vectors or function
pointers, and they end up becoming zeroed.
.It Fl f Ns Cm permissive
Various effects, usually to weaken some conditions.
.Bl -tag -width indent -offset indent
.It with Fl f Ns Cm untyped-nil
Allow local variables named
.Ql nil Ns .
(This will not allow declaring a global of that name.)
.El
.It Fl f Ns Cm variadic-args
Allow variadic parameters to be accessed by QC code. This can be
achieved via the '...' function, which takes a parameter index and a
typename.
.Pp
Example:
.Bd -literal -offset indent
void vafunc(string...count) {
    float i;
    for (i = 0; i < count; ++i)
        print(...(i, string), "\\n");
}
.Ed
.It Fl f Ns Cm legacy-vector-maths
Most Quake VMs, including the one from FTEQW or up till recently
Darkplaces, do not cope well with vector instructions with overlapping
input and output. This option will avoid producing such code.
.It Fl f Ns Cm expressions-for-builtins
Usually builtin-numbers are just immediate constants. With this flag
expressions can be used, as long as they are compile-time constant.
.Pp
Example:
.Bd -literal -offset indent
void printA() = #1; // the usual way
void printB() = #2-1; // with a constant expression
.Ed
.It Fl f Ns Cm return-assignments
Enabiling this option will allow assigning values or expressions to the
return keyword as if it were a local variable of the same type as the
function's signature's return type.
.Pp
Example:
.Bd -literal -offset indent
float bar() { return 1024; }
float fun() {
    return = bar();
    return; // returns value of bar
}
.Ed
.It Fl f Ns Cm unsafe-varargs
When passing on varargs to a different functions, this turns some
static error cases into warnings. Like when the caller's varargs are
restricted to a different type than the callee's parameter. Or a list
of unrestricted varargs is passed into restricted varargs.
.It Fl f Ns Cm typeless-stores
Always use STORE_F, LOAD_F, STOREP_F when accessing scalar variables.
This is somewhat incorrect assembly instruction use, but in all engines
they do exactly the same. This makes disassembly output harder to read,
breaks decompilers, but causes the output file to be better compressible.
.It Fl f Ns Cm sort-operands
In commutative instructions, always put the lower-numbered operand first.
This shaves off 1 byte of entropy from all these instructions, reducing
compressed size of the output file.
.El
.Sh OPTIMIZATIONS
.Bl -tag -width Ds
.It Fl O Ns Cm peephole
Some general peephole optimizations. For instance the code `a = b + c`
typically generates 2 instructions, an ADD and a STORE. This
optimization removes the STORE and lets the ADD write directly into A.
.It Fl O Ns Cm tail-recursion
Tail recursive function calls will be turned into loops to avoid the
overhead of the CALL and RETURN instructions.
.It Fl O Ns Cm overlap-locals
Make all functions which use neither local arrays nor have locals
which are seen as possibly uninitialized use the same local section.
This should be pretty safe compared to other compilers which do not
check for uninitialized values properly. The problem is that there's
QC code out there which really doesn't initialize some values. This is
fine as long as this kind of optimization isn't used, but also, only
as long as the functions cannot be called in a recursive manner. Since
it's hard to know whether or not an array is actually fully
initialized, especially when initializing it via a loop, we assume
functions with arrays to be too dangerous for this optimization.
.It Fl O Ns Cm local-temps
This promotes locally declared variables to "temps". Meaning when a
temporary result of an operation has to be stored somewhere, a local
variable which is not 'alive' at that point can be used to keep the
result. This can reduce the size of the global section.
This will not have declared variables overlap, even if it was
possible.
.It Fl O Ns Cm global-temps
Causes temporary values which do not need to be backed up on a CALL to
not be stored in the function's locals-area. With this, a CALL to a
function may need to back up fewer values and thus execute faster.
.It Fl O Ns Cm strip-constant-names
Don't generate defs for immediate values or even declared constants.
Meaning variables which are implicitly constant or qualified as such
using the 'const' keyword.
.It Fl O Ns Cm overlap-strings
Aggressively reuse strings in the string section. When a string should
be added which is the trailing substring of an already existing
string, the existing string's tail will be returned instead of the new
string being added.
.Pp
For example the following code will only generate 1 string:
.Bd -literal -offset indent
print("Hello you!\\n");
print("you!\\n"); // trailing substring of "Hello you!\\n"
.Ed
.Pp
There's however one limitation. Strings are still processed in order,
so if the above print statements were reversed, this optimization
would not happen.
.It Fl O Ns Cm call-stores
By default, all parameters of a CALL are copied into the
parameter-globals right before the CALL instructions. This is the
easiest and safest way to translate calls, but also adds a lot of
unnecessary copying and unnecessary temporary values. This
optimization makes operations which are used as a parameter evaluate
directly into the parameter-global if that is possible, which is when
there's no other CALL instruction in between.
.It Fl O Ns Cm void-return
Usually an empty RETURN instruction is added to the end of a void
typed function. However, additionally after every function a DONE
instruction is added for several reasons. (For example the qcvm's
disassemble switch uses it to know when the function ends.). This
optimization replaces that last RETURN with DONE rather than adding
the DONE additionally.
.It Fl O Ns Cm vector-components
Because traditional QC code doesn't allow you to access individual
vector components of a computed vector without storing it in a local
first, sometimes people multiply it by a constant like
.Ql '0 1 0'
to get,
in this case, the y component of a vector. This optimization will turn
such a multiplication into a direct component access. If the factor is
anything other than 1, a float-multiplication will be added, which is
still faster than a vector multiplication.
.It Fl O Ns Cm const-fold-dce
For constant expressions that result in dead code (such as a branch whos
condition can be evaluated at compile-time), this will eliminate the branch
and else body (if present) to produce more optimal code.
.El
.Sh CONFIG
The configuration file is similar to regular .ini files. Comments
start with hashtags or semicolons, sections are written in square
brackets and in each section there can be arbitrary many key-value
pairs.
.Pp
There are 3 sections currently:
.Ql flags Ns ,
.Ql warnings Ns ,
.Ql optimizations Ns .
They contain a list of boolean values of the form
.Ql VARNAME = true
or
.Ql VARNAME = false Ns .
The variable names are the same as for the
corresponding
.Fl W , Fl f
or
.Fl O
flag written with only capital letters and
dashes replaced by underscores.
.Pp
Here's an example:
.Bd -literal -offset indent
# a GMQCC configuration file
[flags]
    FTEPP = true
    ADJUST_VECTOR_FIELDS = false
    LNO = true

[warnings]
    UNUSED_VARIABLE = false
    USED_UNINITIALIZED = true

[optimizations]
    PEEPHOLE = true
    TAIL_RECURSION = true
.Ed
.Sh FILES
.Bl -tag -width Ds
.It gmqcc.ini.example
A documented example for a gmqcc.ini file.
.El
.Sh SEE ALSO
.Xr qcvm 1
.Sh AUTHOR
See <http://graphitemaster.github.com/gmqcc>.
.Sh BUGS
Currently the '-fftepp-predefs' flag is not included by '-std=fteqcc',
partially because it is not entirely conformant to fteqcc.
.Pp
Please report bugs on <http://github.com/graphitemaster/gmqcc/issues>,
or see <http://graphitemaster.github.com/gmqcc> on how to contact us.