d0p1/ack
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity
ack/doc/lint/chap4
ceriel c5275f3786 Added
1991-09-30 17:58:00 +00:00

979 lines
29 KiB
Text
Raw Blame History

.NH 1
How lint checks
.NH 2
The first pass first pass data structure
.PP
The data structure of
.I cem
is changed a little and some structures have been added.
.NH 3
The changes
.NH 4
Idf descriptor
.PP
A member
.ft CW
id_line
.R
is added
to the
.I idf
selector.
This line number is used for some warnings.
.NH 4
Def descriptor
.PP
The
.I def
selector is extended with the members
.ft CW
df_set
.R and
df_line.
.R
The
.ft CW
df_used
.R
member did exist already, but was only used for code generation.
This usage is eliminated so it can be used by
.I lint.
The meaning of these members should be clear.
.NH 3
The additions
.NH 4
Lint_stack_entry descriptor
.DS B
.ft CW
struct lint_stack_entry {
struct lint_stack_entry *next;
struct lint_stack_entry *previous;
short ls_class;
int ls_level;
struct state *ls_current;
union {
struct state *S_if;
struct state *S_end;
struct switch_states switch_state;
} ls_states;
};
.R
.DE
.PP
Structure to simulate a stacking mechanism.
.IP \f(CWnext\fP 15
Pointer to the entry on top of this one.
.IP \f(CWprevious\fP
Pointer to the entry beneath this one.
.IP \f(CWls_class\fP
The class of statement this entry belongs to.
Possible classes are \f(CWIF\fP, \f(CWWHILE\fP, \f(CWDO\fP,
\f(CWFOR\fP, \f(CWSWITCH\fP and \f(CWCASE\fP.
.IP \f(CWls_level\fP
The level the corresponding statement is nested.
.IP \f(CWls_current\fP
A pointer to the state descriptor which describes the state
of the function (the state of the automatic variables, if the next
statement can be reached, et cetera) if control passes the
flow of control to the part of the program currently parsed.
The initialization of this state is as follows
.RS
.IP
If \f(CWls_class\fP in [\f(CWIF\fP, \f(CWSWITCH\fP] the state
after parsing the conditional expression.
.IP
If \f(CWls_class\fP in [\f(CWWHILE\fP, \f(CWFOR\fP] the state
after parsing the code between the brackets.
.IP
If \f(CWls_class\fP in [\f(CWDO\fP, \f(CWCASE\fP] the state at
entrance of the statement after the \f(CWDO\fP or \f(CWCASE\fP
token.
.RE
.IP \f(CWls_states\fP 15
Union of pointers to state descriptors containing different information
for different values of \f(CWls_class\fP.
.RS
.IP
If \f(CWls_class\fP is \f(CWIF\fP and in case of parsing an else part,
\f(CWls_states.S_if\fP points to the state that is reached after the
if part.
.IP
If \f(CWls_class\fP in [\f(CWWHILE\fP, \f(CWFOR\fP, \f(CWDO\fP]
then \f(CWls_states.S_end\fP contains a conservative description
of the state of the program after `jumping'
to the end of the statement after the \f(CWWHILE\fP, \f(CWDO\fP
or \f(CWFOR\fP token.
I.e. the state at reaching a break (not inside a switch) or
continue statement.
.IP
If ls_class is \f(CWSWITCH\fP, \f(CWls_states\fP is used as a structure
.DS B
.ft CW
struct switch_states {
struct state S_case;
struct state S_break;
};
.R
.DE
containing two pointers to state descriptors.
\f(CWls_states.switch_state.S_case\fP contains
a conservative description
of the state of the program after \f(CWcase ... case\fP
parts are parsed.
\f(CWls_states.switch_state.S_break\fP the state after parsing
all the \f(CWcase ... break\fP parts.
The reason for \f(CWls_states.switch_state.default_met\fP should be
self-explanatory.
.IP
In case \f(CWls_class\fP is \f(CWCASE\fP, \f(CWls_states\fP is not used.
.RE
.NH 4
State descriptor
.DS B
.ft CW
struct state {
struct state *next;
struct auto_def *st_auto_list;
int st_nrchd;
int st_warned;
};
.R
.DE
.IP \f(CWst_auto_list\fP 15
Pointer to a list of definitions of the automatic variables whose
scope contain the current position in the program.
.IP \f(CWst_nrchd\fP
True if the next statement can't be reached.
.IP \f(CWst_warned\fP
True if a warning has already been given.
.NH 4
Auto_def descriptor
.DS B
.ft CW
struct auto_def {
struct auto_def *next;
struct idf *ad_idf;
struct def *ad_def;
int ad_used;
int ad_set;
int ad_maybe_set;
};
.R
.DE
.IP \f(CWnext\fP 15
Points to the next auto_definition of the list.
.IP \f(CWad_idf\fP
Pointer to the idf descriptor associated with this auto_definition.
.IP \f(CWad_def\fP
Ditto for def descriptor.
.IP \f(CWad_used\fP
Indicates the state of this automatic variable.
Ditto for \f(CWad_set\fP and \f(CWad_maybe_set\fP.
Only one of \f(CWad_set\fP and \f(CWad_maybe_set\fP may be true.
.NH 4
Expr_state descriptor
.DS B
.ft CW
struct expr_state {
struct expr_state *next;
struct idf *es_idf;
arith es_offset;
int es_used;
int es_set;
};
.R
.DE
.PP
This structure is introduced to keep track of which variables,
array entries and structure members (union members) are set
and/or used in evaluating an expression.
.IP \f(CWnext\fP 15
Pointer to the next descriptor of this list.
.IP \f(CWes_idf\fP
Pointer to the idf descriptor this descriptor belongs to.
.IP \f(CWes_offset\fP
In case of an array, a structure or union, this member contains
the offset the compiler would generate for locating the array
entry or structure/union member.
.IP \f(CWes_used\fP
True if the indicated memory location is used in evaluating the
expression.
.IP \f(CWes_set\fP
Ditto for set.
.NH 4
Outdef descriptor
.DS B
.ft CW
struct outdef {
int od_class;
char *od_name;
char *od_file;
unsigned int od_line;
int od_nrargs;
struct tp_entry *od_entry;
int od_returns;
struct type *od_type;
};
.DE
.R
.PP
As structures of this type are not allocated dynamically by a
storage allocator, it contains no next member.
An outdef can be given to to \f(CWoutput_def()\fP to be passed to the
second pass.
Basically this forms the interface with the second pass.
.IP \f(CWod_class\fP 15
Indicates what kind of definition it is.
Possible classes are \f(CWEFDF\fP, \f(CWEVDF\fP, \f(CWSFDF\fP,
\f(CWSVDF\fP, \f(CWLFDF\fP, \f(CWLVDF\fP,
\f(CWEFDC\fP, \f(CWEVDC\fP, \f(CWIFDC\fP, \f(CWFC\fP, \f(CWVU\fP.
([\f(CWE\fPxternal, \f(CWS\fPtatic, \f(CWL\fPibrary, \f(CWI\fPmplicit]
[\f(CWF\fPunction, \f(CWV\fPariable]
[\f(CWD\fPe\f(CWF\fPinition, \f(CWD\fPe\f(CWC\fPlaration,
\f(CWC\fPall, \f(CWU\fPsage])
.IP \f(CWod_name\fP
The name of the function or variable.
.IP \f(CWod_file\fP
The file this definition comes from.
.IP \f(CWod_nrargs\fP
If \f(CWod_class\fP is one of \f(CWEFDF\fP, \f(CWSFDF\fP or
\f(CWLFDF\fP, this member contains the
number of arguments this function has.
If the function was preceded by the pseudocomment
\f(CW/*\ VARARGS\ */\fP,
\f(CWod_nrargs\fP gets the value \f(CW-1-n\fP.
.IP \f(CWod_entry\fP
A pointer to a list of \f(CWod_nrargs\fP cells, each containing a
pointer to the type descriptor of an argument. (\f(CW-1-od_nrargs\fP
cells if
\f(CWod_nrargs < 0\fP.)
\f(CWTp_entry\fP is defined as
.DS B
.ft CW
struct tp_entry {
struct tp_entry *next; /* pointer to next cell */
struct type *te_type; /* an argument type */
};
.R
.DE
.IP \f(CWod_returns\fP 15
For classes \f(CWEFDF\fP, \f(CWSFDF\fP and \f(CWLFDF\fP this
member tells if the function returns an expression or not.
In case \f(CWod_class\fP is \f(CWFC\fP it is true if the value
of the function is used, false otherwise.
For other classes this member is not used.
.IP \f(CWod_type\fP
A pointer to the type of the function or variable defined or
declared.
Not used for classes \f(CWFC\fP and \f(CWVU\fP.
.NH 2
The first pass checking mechanism
.PP
In the description of the implementation of the pass one
warnings, it is assumed that the reader is familiar with the
\fILLgen\fP parser generator, as described in [6].
.NH 3
Used and/or set variables
.PP
To be able to give warnings like
.ft CW
%s used before set
.R
and
.ft CW
%s set but not used in function %s
.R
, there needs to be a way to keep track of the state of a variable.
A first approach to do this was by adding two fields to the
\fIdef\fP selector:
.ft CW
df_set
.R
and
.ft CW
df_used.
.R
While parsing the program, each time an expression was met
this expression was analyzed and the fields of each \fIdef\fP
selector were possibly set during this analysis.
This analysis was done by passing each expression to a
function
.ft CW
lint_expr
.R
, which walks the expression tree in a way similar to the function
\f(CWEVAL\fP in the file \fIeval.c\fP of the original
.I
cem
.R
compiler.
This approach has one big disadvantage: it is impossible to keep
track of the flow of control of the program.
No warning will be given for the program fragment of figure 3.
.KF
.DS B
.ft CW
func()
{
int i;
if (cond)
i = 0;
else
use(i); /* i may be used before set */
}
.I
.DE
.br
.ce
figure\ 3.
.R
.KE
.PP
It is clear that it would be nice having
.I lint
warn for this construction.
.PP
This was done in the second approach.
When there was a choice between two statements, each statement
was parsed with its own copy of the state at entrance of the
.I
choosing statement.
.R
A state consisted of the state of the automatic variables
(including register variables).
In addition to the possibilities of being used and set,
a variable could be \fImaybe set\fP.
These states were passed between the statement parsing routines
using the \fILLgen\fP parameter mechanism.
At the end of a choosing statement, the two states were merged
into one state, which became the state after this statement.
The construction of figure 4 was now detected, but switch
statements still gave problems and continue and break statements
were not understood.
The main problem of a switch statement is, that the closing bracket
(`\f(CW)\fP') has to be followed by a \fIstatement\fP.
The syntax shows no choice of statements, as is the case with
if, while, do and for statements.
Using the \fILLgen\fP parameter mechanism, it is not a trivial
task to parse the different case parts of a switch statement
with the same initial state and to merge the results into one
state.
This observation led to the third and final approach, as described
next.
.PP
Instead of passing the state of the program through the statements
parsing routines using the \fILLgen\fP parameters, a special stack is
introduced, the
.I lint_stack.
When a choosing statement is parsed, an entry is pushed on the stack
containing the information that is needed to keep track of the
state of the program.
Each entry contains a description of the
.I current
state of the program and a field that indicates what part of the
program the parser is currently parsing.
For all the possible choosing statements I describe the actions
to be taken.
.PP
At entrance of an if statement, an entry is pushed on the stack
with the current state being a copy of the current state of the
stack element one below.
The class of this entry is \f(CWIF\fP.
At reaching the else part, the current state is moved to
another place in this stack entry (to \f(CWS_IF\fP), and a new copy
of the current state at entrance of this if statement is made.
At the end of the else part, the two states are merged into
one state, the new current state, and the \f(CWIF\fP entry is
popped from the stack.
If there is no else part, then the state that is reached after
parsing the if part is merged with the current state at entrance
of the if statement into the new current state.
.PP
At entrance of a while statement a \f(CWWHILE\fP entry is pushed
on the stack containing a copy of the current state.
If a continue or break statement is met in the while statement,
the state at reaching this continue or break statement is
merged with a special state in the \f(CWWHILE\fP entry, called
\f(CWS_END\fP.
(If \f(CWS_END\fP did not yet contain a state, the state is copied
to \f(CWS_END\fP.)
At the end of the while statement this \f(CWS_END\fP is merged with the
current state, which result is merged with the state at entrance
of the while statement into the new current state.
.PP
A for statement is treated similarly.
A do statement is treated the same way too, except that \f(CWS_END\fP
isn't merged with the state at entrance of the do statement,
but becomes the new current state.
.PP
For switch statements a \f(CWSWITCH\fP entry is pushed on the stack.
Apart from the current state, this entry contains two other
states, \f(CWS_BREAK\fP and \f(CWS_CASE\fP.
\f(CWS_BREAK\fP initially contains no state, \f(CWS_CASE\fP
initially contains a
copy of the current state at entrance of the switch statement.
After parsing a case label, a \f(CWCASE\fP entry is pushed on the stack,
containing a copy of the current state.
If, after zero or more statements, we meet another case label,
the state at reaching this case label is merged with \f(CWS_CASE\fP
of the \f(CWSWITCH\fP entry below and a new copy of the state
at entrance
of the switch statement is put in the \f(CWCASE\fP entry.
If we meet a break statement, we merge the current state with
\f(CWS_BREAK\fP of the \f(CWSWITCH\fP entry below and pop the
\f(CWCASE\fP entry.
In addition to this, the occurrence of a default statement
inside the switch statement is recorded in the \f(CWSWITCH\fP entry.
At the end of the switch statement we check if we have met a
default statement.
If not, \f(CWS_BREAK\fP is merged with the current state at entrance
of the switch statement. (Because it is possible that no case
label will be chosen.)
Next the \f(CWS_CASE\fP is `special_merged' with \f(CWS_BREAK\fP
into the new current state.
For more details about these merge functions see the sources.
.PP
With the approach described above,
.I lint
is aware of the flow
of control in the program.
There still are some doubtful constructions
.I lint
will not detect and there are some constructions (although rare)
for which
.I lint
gives an incorrect warning (see figure 4).
.KF
.DS B
.ft CW
{
int i;
for (;;) {
if (cond) {
i = 0;
break;
}
}
use(i);
/* lint warns: maybe i used before set
* although the fragment is correct
*/
}
.DE
.br
.I
.ce
figure\ 4.
.R
.KE
.PP
A nice advantage of the method is, that the parser stays clear,
i.e. it isn't extended with extra parameters which must pass the
states.
In this way the parser still is very readable and we have a nice
interface with
.I lint
using function calls.
.NH 3
Undefined evaluation orders
.PP
In expressions the values of some variables are used and some
variables are set.
Of course, the same holds for subexpressions.
The compiler is allowed to choose the order of evaluation of
subexpressions involving a commutative and associative operator
(\f(CW*\fP, \f(CW+\fP, \f(CW&\fP, \f(CW|\fP, \f(CW^\fP),
the comma in a parameter list or an assignment operator.
In section 3.4 it is made clear that this will lead to
statements with ambiguous semantics.
.PP
The way these constructs are detected is rather straight forward.
The function which parses an expression (\f(CWlint_expr\fP)
returns a linked
list containing information telling which variables are set and
which variables are used.
A variable is indicated by its
.I idf
descriptor and an
.I offset.
This offset is needed for discriminating entries of the same
array and members of the same structure or union, so it is
possible to warn about the statement
.ft CW
a[b[0]]\ =\ b[0]++;.
.R
When \f(CWlint_expr\fP meets a commutative operator (with respect to the
evaluation order), it calls itself recursively with the operands
of the operator as expression.
The returned results are checked for undefined evaluation orders
and are put together.
This is done by the function \f(CWcheck_and_merge\fP.
.NH 3
Useless statements
.PP
Statements which compute a value that is not used,
are said to have a \fInull effect\fP.
Examples are \f(CWx = 2, 3;\fP, \f(CWf() + g();\fP and
\f(CW*p++;\fP.
(\f(CW*\fP and \f(CW++\fP have the same precedence and associate
from right to left.)
.PP
A conditional expression computes a value too.
If this value isn't used, it is better to use an if-else
statement.
So, if
.I lint
sees
.DS B
.ft CW
b ? f() : g();
.R
.DE
.LP
it warns \f(CWuse if-else construction\fP.
.NH 3
Not-reachable statements
.PP
The algorithm to detect not-reachable statements (including not
reachable initializations) is as follows.
Statements after a label and a case statement and the compound
statement of a function are always reachable.
Other statements are not-reachable after:
.QS
.RS
.IP - 1
a goto statement
.IP -
a return statement
.IP -
a break statement
.IP -
a continue statement
.IP -
a switch statement
.IP -
an endless loop (a while, do or for loop with a conditional
which always evaluates to true and without a break statement)
.IP -
an if-else statement of which both if part and else part
end up in a not-reachable state
.IP -
a switch statement of which all \f(CWcase ... break\fP parts
(including
a \f(CWdefault ... break\fP part) end up in a not-reachable state
.IP -
the pseudocomment \f(CW/*\ NOTREACHED\ */\fP
.RE
.QE
.PP
The algorithm is easily implemented using the \f(CWst_nrchd\fP selector
in the
.I state
descriptor.
The \f(CWst_warned\fP selector is used to prevent superfluous warnings.
To detect an endless loop, after a while (<true>), for (..;<true>;..)
and do part the current state of the stack entry beneath the top one
is set to not reached.
If, in the statement following, a break statement is met, this same
state is set to reached.
If the while (<cond>) part of the do statement is met, this state
is set to reached if <cond> doesn't evaluates to true.
The detection of not-reachable statements after a switch statement
is done in a similar way.
In addition it is checked if a default statement isn't met, in
which case the statement after the switch statement can be reached.
The warning \f(CWstatement not reached\fP is not given for compound
statements.
If
.I lint
did, it would warn for the compound statement in a switch statement,
which would be incorrect.
.PP
Not-reachable statements are still interpreted by
.I lint.
I.e. when
.I lint
warns that some statement can't be reached, it assumes this is
not what the programmer really wants and it ignores this fact.
In this way a lot of useless warnings are prevented in the case of
a not-reachable statement.
See figure 5.
.KF
.DS B
.ft CW
{
int i;
for (;;) {
/* A loop in which the programmer
* forgot to introduce a conditional
* break statement.
* Suppose i is not used in this part.
*/
}
/* some more code in which i is used */
}
/* The warning "statement not reached" highlights the bug.
* An additional warning "i unused in function %s" is
* formally correct, but doesn't provide the programmer
* with useful information.
*/
.DE
.I
.ce
figure\ 5.
.R
.KE
.NH 3
Functions returning expressions and just returning
.PP
Each time a return statement is met,
.I lint
checks if an expression is returned or not.
If a function has a return with expression and a return without
expression,
.I lint
warns
.ft CW
function %s has return(e); and return;.
.R
If the flow of control can
.I
fall through
.R
the end of the compound statement of a function, this indicates
an implicit return statement without an expression.
If the end of the compound statement of the function can be reached,
.I lint
introduces this implicit return statement without expression.
.PP
Sometimes the programmer knows for sure that all case parts inside
a switch statement include all possible cases, so he doesn't
introduce a default statement.
This can lead to an incorrect warning.
Figure 6 shows how to prevent this warning.
.KF
.DS B
.ft CW
func()
{
switch (cond) {
case 0: return(e0);
case 1: return(e1);
}
/* NOTREACHED */
}
/* no warning: "function func has return(e); and return; */
.DE
.I
.ce
figure\ 6.
.R
.KE
.PP
The pseudocomment \f(CW/*\ NOTREACHED\ */\fP can also be used to tell
.I lint
that some function doesn't return. See figure 7.
.KS
.DS B
.ft CW
func()
{
switch (cond) {
case 0: return(e0);
case 1: return(e1);
default: error(); /* calls exit or abort */
/* NOTREACHED */
}
}
/* no warning: "function func has return(e); and return;" */
.I
.DE
.ce
figure\ 7.
.R
.KE
.NH 3
Output definitions for the second pass
.PP
The first pass can only process one program file.
To be able to process a program that spreads over more than one file,
the first pass outputs definitions that are processed by a second
pass.
The format of such a definition is different for different classes:
.PP
For class in {EFDF, SFDF, LFDF}
.DS C
<name>:<class>:<file>:<line>:<nr of args>:<type of args>:<returns value>:<type>
.DE
.LP
A negative \fInr of args\fP indicates that the function can be called with
a varying number of arguments.
.PP
For class = FC
.DS C
<name>:<class>:<file>:<line>:<value is used>:<type>
.DE
.LP
The \fIvalue is used\fP part can have three meanings:
the value of the function is ignored;
the value of the function is used;
the value of the function is cast to type \fIvoid\fP.
.PP
For other classes
.DS C
<name>:<class>:<file>:<line>:<type>
.DE
.LP
Definitions of class VU (Variable Usage) are only output for \fIused\fP
global variables.
.PP
Structure and union types that are output to the intermediate file
are simplified.
(The following occurrences of \fIstructure\fP should be
read as \fIstructure or union\fP and \fIstruct\fP as \fIstruct or
union\fP.)
Structures that are identified by a \fIstructure tag\fP are output
to the intermediate file as \f(CWstruct <tag>\fP.
Structures without a structure tag are output as
\f(CWstruct {<mems>}\fP with \f(CW<mems>\fP a semicolon-separated
list of types of the members of this structure.
An alternative way would be to output the complete structure definition.
However, this gives practical problems.
It is allowed to define some object of a structure type with a
structure tag, without this structure being defined at that place.
The first approach leaves errors, such as in figure 8, undetected.
.KF
.DS B
.ft CW
"a.c" "b.c"
struct str { struct str {
float f; int i;
} s; };
main() func(s)
{ struct str s;
func(s); {}
}
.I
.DE
.ce
figure\ 8.
.R
.KE
.PP
To be able to detect these errors, the first pass should also output
definitions of structure tags.
The example of figure 8 would then get a warning like
.ft CW
structure str defined inconsistently
.R
.PP
More information on these definitions in section 4.3 and 4.4.
.NH 3
Generating libraries
.PP
.I Lint
knows the library `-lc', `-lm' and `-lcurses'.
If a program uses some other library, it is possible to generate
a corresponding \fIlint library\fP.
To do this, precede all the C source files of this library by
the pseudocomment \f(CW/*\ LINTLIBRARY\ */\fP.
Then feed these files one by one to the first pass of
.I lint
collecting the standard output in a file and ignoring the warnings.
The resulting file contains library definitions of the functions
and external variables defined in the library sources, and not more
than that.
If this file is called `llib-l\fIname\fP.ln
.I lint
can be told to search the library by passing it as argument in
the command line `-llib-l\fIname\fP.ln.
The implementation of this feature is simple.
.PP
As soon as the pseudocomment \f(CW/*\ LINTLIBRARY\ */\fP is met,
only function and variable definitions are output with class LFDF
and LVDF respectively.
Other definitions, which otherwise would have been output, are
discarded.
.PP
Instead of generating a special lint library file, one can make a
file containing the library definitions and starting with
\f(CW/* LINTLIBRARY */\fP.
This file can then be passed to
.I lint
just by its name.
This method isn't as efficient as the first one.
.NH 3
Interpreting the pseudocomments
.PP
The interpretation of the pseudocomments is done by the lexical
analyzer, because this part of the program already took care of the
comments.
At first sight this seems very easy: as soon as some pseudocomment
is met, raise the corresponding flag.
Unfortunately this doesn't work.
The lexical analyzer is a \fIone token look ahead scanner\fP.
This causes the above procedure to raise the flags one token too
soon.
A solution to get the right effect is to reserve two flags per
pseudocomment.
The first is set as soon as the corresponding pseudocomment is
scanned.
At the returning of each token this flag is moved to the second flag.
The delay in this way achieved makes the pseudocomments have effect
at the correct place.
.NH 2
The second pass data structure
.NH 3
Inp_def descriptor
.DS B
.ft CW
struct inp_def {
struct inp_def *next;
int id_class;
char id_name[NAMESIZE];
char id_file[FNAMESIZE];
unsigned int id_line;
int id_nrargs;
char argtps[ARGSTPSSIZE];
int id_returns;
char id_type[TYPESIZE];
int id_called;
int id_used;
int id_ignored;
int id_voided;
};
.R
.DE
.PP
This description is almost similar to the \fIoutdef\fP descriptor as
described in 4.1.2.5.
There are some differences too.
.IP \f(CWnext\fP 15
As structures of this type are allocated dynamically, this field
is added so the same memory allocator as used in the first pass can be
used.
.LP
\f(CWid_called
.br
id_used
.br
id_ignored\fP
.IP \f(CWid_voided\fP 15
Some additional fields only used for function definitions.Their
meaning should be clear.
.PP
The other fields have the same meaning as the corresponding fields
in the \fIoutdef\fP descriptor.
Some attention should be paid to \f(CWid_argtps\fP and \f(CWid_type\fP.
These members have type \f(CWarray of char\fP, in contrast to
their counterparts in the \fIoutdef\fP descriptor.
The only operation performed on types is a check on equality.
Types are output by the first pass as a string describing the type.
The type of \f(CWi\fP in \f(CWint *i();\fP e.g. is output as
\f(CWint *()\fP.
Such a string is best put in an \f(CWarray of char\fP to be compared
easily.
.NH 2
The second pass checking mechanism
.PP
After all the definitions that are output by the first pass are
sorted by name, the definitions belonging to one name are ordered
as follows.
.QS
.RS
.IP - 1
external definitions
.IP -
static definitions
.IP -
library definitions
.IP -
declarations
.IP -
function calls
.IP -
variable usages
.RE
.QE
.PP
The main program of the second pass is easily explained.
For all different names, do the following.
First read the definitions.
If there is more than one definition, check for conflicts.
Then read the declarations, function calls and variable usages and
check them against the definitions.
After having processed all the declarations, function calls and
variable usages, check the definitions to see if they are used
correctly.
The next three paragraphs will explain the three most important
functions of the program.
.NH 3
Read_defs()
.PP
This function reads all definitions belonging to the same name.
Only one external definition is allowed, so if there are more, a
warning is given.
In different files it is allowed to define static functions or
variables with the same name.
So if a static function is read, \f(CWread_defs\fP checks if there isn't
already an external definition, and if not it puts the static
definition in the list of static definitions, to be used later.
If no external or static definitions are met, a library definition is
taken as definition.
If a function or a variable is defined with the same name as a function
or a variable in a library (which is allowed)
.I lint
gives a warning.
Of course it is also possible that there is no definition at all.
In that case \f(CWcheck\fP will warn.
.NH 3
Check()
.PP
\f(CWCheck\fP verifies declarations, function calls and variable
usages against the definitions.
For each of these entries the corresponding definition is looked up.
As there may be more than one static definition, first a static
definition from the same file as the entry is searched.
If not present, the external definition (which may be a library
definition) is taken as definition.
If no definition can be found and the current entry is an external
declaration,
.I lint
warns.
However in the case of an implicit function declaration
.I lint
will not warn, because
we will get a warning \f(CW%s used but not defined\fP later on.
Next a check is done if the declarations are consistent with their
definitions.
After the declarations, the function calls and variable usages are
verified against their corresponding definitions.
If no definition exists,
.I lint
warns.
Else the field \f(CWid_called\fP is set to 1.
(For variable definitions this should be interpreted as \fIused\fP.)
For variable usages this will be all.
If we are processing a function call we also check the number and types
of the arguments and we warn for function values which are used from
functions that don't return a value.
For each function call we administrate if a function value is used,
ignored or voided.
.NH 3
Check_usage()
.PP
Checks if the external definition and static definitions are used
correctly.
If a function or variable is defined but never used,
.I lint
warns, except for library definitions.
Functions, which return a value but whose value is always or
sometimes ignored, get a warning.
(A function value which is voided (cast to void) is not ignored,
but it isn't used either.)
.bp
Reference in a new issue View git blame Copy permalink