I’ve been thinking about code size, bytecode formats, and so on: what kind of tradeoffs in program representation are best? Existing microbenchmarks tend to be very atypical of real code; real code is mostly much more straightforward and boring than microbenchmarks.
So here are some sample code fragments to try different compilation strategies with (after hand-translating them to some uniform syntax, I suppose). These are chosen to cover the range of real code found in the wild, although they are biased towards code bases of high quality and towards medium-length functions: functions between 8 and 64 lines are better represented. There is code here in Pascal, C, C++, C#, Java, Golang, Perl, Tcl, Python, PHP, JS, Lua, Elisp, Ruby, Haskell, R, and Visual Basic — every popular programming language except Excel, Swift, assembly, SQL, and Matlab. The idea here is to include things written in somewhat different paradigms.
I thought about including code in bash, TeX, assembly, SQL, CSS, Prolog, and HTML, but their execution models are too different from the mainstream to be reasonably translatable to a mainstream virtual machine. Even R and Haskell are kind of pushing it.
The classic dumb Fibonacci, in C:
fib(n) { return n < 2 ? 1 : fib(n-1) + fib(n-2); }
A nearly totally vacuous class definition:
class Match:
def __init__(self, length, rendering):
self.length = length
self.rendering = rendering
A test of the mod_pubsub JS interface, some code I think I wrote in 2000, unless it was Ben or Rohit:
function do_it()
{
var topic = kn_argv['kn_topic'];
kn_publish(topic, {}, { onSuccess: function(e) {succeed()},
onError: function(e) {fail(e.kn_payload)} });
}
Part of the Perl abstract base class DBD::File:
sub table_meta_attr_changed
{
my ($class, $meta, $attrib, $value) = @_;
defined $reset_on_modify{$attrib} and
delete $meta->{$reset_on_modify{$attrib}} and
$meta->{initialized} = 0;
} # table_meta_attr_changed
The Lua code in LuaTeX for computing the full platform string on Linux:
function os.resolvers.platform(t,k)
local platform = "linux"
os.setenv("MTX_PLATFORM",platform)
os.platform = platform
return platform
end
A byteswapping function by Jon Lech Johansen in C#:
byte [] NetToHost( byte [] Input, int Pos, int Count )
{
if( BitConverter.IsLittleEndian )
{
for( int i = 0; i < Count; i++ )
{
Array.Reverse( Input, Pos + (i * 4), 4 );
}
}
return Input;
}
In GNU cp, some slight options parsing hackery:
/* For long options that have no equivalent short option, use a
non-character as a pseudo short option, starting with CHAR_MAX + 1. */
enum
{
ATTRIBUTES_ONLY_OPTION = CHAR_MAX + 1,
COPY_CONTENTS_OPTION,
NO_PRESERVE_ATTRIBUTES_OPTION,
PARENTS_OPTION,
PRESERVE_ATTRIBUTES_OPTION,
REFLINK_OPTION,
SPARSE_OPTION,
STRIP_TRAILING_SLASHES_OPTION,
UNLINK_DEST_BEFORE_OPENING
};
A function from netqmail to downcase a NUL-terminated string:
#include "case.h"
void case_lowers(s)
char *s;
{
unsigned char x;
while (x = *s) {
x -= 'A';
if (x <= 'Z' - 'A') *s = x + 'a';
++s;
}
}
Part of Jon Lech Johansen’s DeDRMS utility in C#:
class M4PStream
{
// ...
public M4PStream( FileStream fs )
{
br = new BinaryReader( fs );
bw = new BinaryWriter( fs );
sbuffer = br.ReadBytes( Convert.ToInt32( fs.Length ) );
alg = Rijndael.Create();
alg.Mode = CipherMode.CBC;
alg.Padding = PaddingMode.None;
}
// ...
}
A function from the Nouveau video driver in the Linux kernel:
u16
nvbios_cstepEe(struct nvkm_bios *bios, int idx, u8 *ver, u8 *hdr)
{
u8 cnt, len, xnr, xsz;
u16 data = nvbios_cstepTe(bios, ver, hdr, &cnt, &len, &xnr, &xsz);
if (data && idx < cnt) {
data = data + *hdr + (idx * len);
*hdr = len;
return data;
}
return 0x0000;
}
Part of the Elisp code for Flymake, which gives you on-the-fly error checking in Emacs, in this case converting between a couple of slightly different data formats in order to pop up a list of suggested corrections when you click on an error:
(defun flymake-make-emacs-menu (menu-data)
"Return a menu specifier using MENU-DATA.
MENU-DATA is a list of error and warning messages returned by
`flymake-make-err-menu-data'.
See `x-popup-menu' for the menu specifier format."
(let* ((menu-title (nth 0 menu-data))
(menu-items (nth 1 menu-data))
(menu-commands (mapcar (lambda (foo)
(cons (nth 0 foo) (nth 1 foo)))
menu-items)))
(list menu-title (cons "" menu-commands))))
The SafeBuffer indexing method from Ruby ActiveSupport:
def [](*args)
return super if args.size < 2
if html_safe?
new_safe_buffer = super
new_safe_buffer.instance_eval { @html_safe = true }
new_safe_buffer
else
to_str[*args]
end
end
A function from the Linux network scheduling code:
static int ingress_dump(struct Qdisc *sch, struct sk_buff *skb)
{
struct nlattr *nest;
nest = nla_nest_start(skb, TCA_OPTIONS);
if (nest == NULL)
goto nla_put_failure;
return nla_nest_end(skb, nest);
nla_put_failure:
nla_nest_cancel(skb, nest);
return -1;
}
A function from Godoc, the Golang documentation extractor:
func findSpec(list []ast.Spec, id *ast.Ident) ast.Spec {
for _, spec := range list {
switch s := spec.(type) {
case *ast.ImportSpec:
if s.Name == id {
return s
}
case *ast.ValueSpec:
for _, n := range s.Names {
if n == id {
return s
}
}
case *ast.TypeSpec:
if s.Name == id {
return s
}
}
}
return nil
}
Part of the GNU ISO C++ library implementing the thin_heap type:
PB_DS_CLASS_T_DEC
inline void
PB_DS_CLASS_C_DEC::
fix_sibling_rank_1_unmarked(node_pointer p_y)
{
_GLIBCXX_DEBUG_ASSERT(p_y->m_p_prev_or_parent != 0);
_GLIBCXX_DEBUG_ONLY(node_pointer p_w = p_y->m_p_l_child;)
_GLIBCXX_DEBUG_ASSERT(p_w != 0);
_GLIBCXX_DEBUG_ASSERT(p_w->m_p_next_sibling == 0);
_GLIBCXX_DEBUG_ASSERT(p_y->m_p_next_sibling == 0);
p_y->m_p_next_sibling = p_y->m_p_l_child;
p_y->m_p_next_sibling->m_p_prev_or_parent = p_y;
p_y->m_p_l_child = 0;
PB_DS_ASSERT_NODE_CONSISTENT(p_y, false)
}
Part of the tests for d3.js:
"can output a percentage with rounding and sign": function(format) {
var f = format("+.2p");
assert.strictEqual(f(.00123), "+0.12%");
assert.strictEqual(f(.0123), "+1.2%");
assert.strictEqual(f(.123), "+12%");
assert.strictEqual(f(1.23), "+120%");
assert.strictEqual(f(-.00123), "-0.12%");
assert.strictEqual(f(-.0123), "-1.2%");
assert.strictEqual(f(-.123), "-12%");
assert.strictEqual(f(-1.23), "-120%");
},
Another function from netqmail, this one for comparing two strings lexically:
int str_diffn(s,t,len)
register char *s;
register char *t;
unsigned int len;
{
register char x;
for (;;) {
if (!len--) return 0; x = *s; if (x != *t) break; if (!x) break; ++s; ++t;
if (!len--) return 0; x = *s; if (x != *t) break; if (!x) break; ++s; ++t;
if (!len--) return 0; x = *s; if (x != *t) break; if (!x) break; ++s; ++t;
if (!len--) return 0; x = *s; if (x != *t) break; if (!x) break; ++s; ++t;
}
return ((int)(unsigned int)(unsigned char) x)
- ((int)(unsigned int)(unsigned char) *t);
}
A module from the Free Pascal compiler providing access to an AmigaOS timer function:
unit timerutils;
interface
uses exec, timer, amigalib;
Function CreateTimer(theUnit : longint) : pTimeRequest;
{ some stuff omitted }
implementation
Function CreateTimer(theUnit : longint) : pTimeRequest;
var
Error : longint;
TimerPort : pMsgPort;
TimeReq : pTimeRequest;
begin
TimerPort := CreatePort(Nil, 0);
if TimerPort = Nil then
CreateTimer := Nil;
TimeReq := pTimeRequest(CreateExtIO(TimerPort,sizeof(tTimeRequest)));
if TimeReq = Nil then begin
DeletePort(TimerPort);
CreateTimer := Nil;
end;
Error := OpenDevice(TIMERNAME, theUnit, pIORequest(TimeReq), 0);
if Error <> 0 then begin
DeleteExtIO(pIORequest(TimeReq));
DeletePort(TimerPort);
CreateTimer := Nil;
end;
TimerBase := pointer(TimeReq^.tr_Node.io_Device);
CreateTimer := pTimeRequest(TimeReq);
end;
{ lots of stuff omitted }
end.
From quodlibet, some PyGTK code with a callback:
def edit_patterns(cls, button):
def valid_uri(s):
# TODO: some pattern validation too (that isn't slow)
try:
p = Pattern(s)
u = URI(s)
return (p and u.netloc and
u.scheme in ["http", "https", "ftp", "file"])
except ValueError:
return False
win = StandaloneEditor(filename=cls.PATTERNS_FILE,
title=_("Search URL patterns"), initial=cls.DEFAULT_URL_PATS,
validator=valid_uri)
win.show()
From slide-rule.tcl, some Tk code with callbacks that I wrote in like 1997:
proc sl_ru_bind {canvas} {
bind $canvas <1> {
global old_x
set old_x %x
global dragobj
set dragobj [groupof %W [%W find closest %x %y]]
}
bind $canvas <B1-Motion> {
global old_x dragobj
%W move $dragobj [expr (%x - $old_x) / 2] 0
set old_x %x
}
}
A perspective projection in JS from some code I wrote to design domes:
function project(points, zDist, zMin, width) {
var rv = [];
for (var ii = 0; ii < points.length; ii++) {
var p = points[ii]
, xx = p[0]
, yy = p[1]
, zz = p[2] + zDist
;
if (zz > zMin) {
rv.push([width * xx / zz, width * yy / zz]);
} else {
rv.push([NaN, NaN]);
}
}
return rv;
}
Some disappointingly boilerplatish code from Lucene:
public String toString(String field) {
StringBuffer buffer = new StringBuffer();
buffer.append("spanNot(");
buffer.append(include.toString(field));
buffer.append(", ");
buffer.append(exclude.toString(field));
buffer.append(")");
buffer.append(ToStringUtils.boost(getBoost()));
return buffer.toString();
}
From Darius’s constraint-programming draft in JS:
// Try to reduce eqns to an equivalent system with each variable
// defined by a single equation. (The result may be inconsistent
// or underconstrained.)
function reduceEquations(eqns) {
for (let i = 0; i < eqns.length; ++i) {
const eqi = eqns[i];
const v = eqi.aVariable();
if (v === null) continue;
for (let j = 0; j < i; ++j) {
eqns[j] = eqns[j].substituteFor(v, eqi);
if (eqns[j].isInconsistent()) {
return {isConsistent: false};
}
}
};
return {isConsistent: true,
equations: (eqns.filter(eqn => !eqn.isTautology())
.map(eqn => eqn.normalize()))};
}
The glue code in the Lua 5.1 OS interface to allow Lua code to call mktime(3):
static int os_time (lua_State *L) {
time_t t;
if (lua_isnoneornil(L, 1)) /* called without args? */
t = time(NULL); /* get current time */
else {
struct tm ts;
luaL_checktype(L, 1, LUA_TTABLE);
lua_settop(L, 1); /* make sure table is at the top */
ts.tm_sec = getfield(L, "sec", 0);
ts.tm_min = getfield(L, "min", 0);
ts.tm_hour = getfield(L, "hour", 12);
ts.tm_mday = getfield(L, "day", -1);
ts.tm_mon = getfield(L, "month", -1) - 1;
ts.tm_year = getfield(L, "year", -1) - 1900;
ts.tm_isdst = getboolfield(L, "isdst");
t = mktime(&ts);
}
if (t == (time_t)(-1))
lua_pushnil(L);
else
lua_pushnumber(L, (lua_Number)t);
return 1;
}
A fairly boilerplatish class definition in Perl:
package Pod::Simple::PullParserToken;
# Base class for tokens gotten from Pod::Simple::PullParser's $parser->get_token
@ISA = ();
$VERSION = '3.16';
use strict;
sub new { # Class->new('type', stuff...); ## Overridden in derived classes anyway
my $class = shift;
return bless [@_], ref($class) || $class;
}
sub type { $_[0][0] } # Can't change the type of an object
sub dump { Pod::Simple::pretty( [ @{ $_[0] } ] ) }
sub is_start { $_[0][0] eq 'start' }
sub is_end { $_[0][0] eq 'end' }
sub is_text { $_[0][0] eq 'text' }
1;
__END__
Some R code from the CRAN survival analysis package, doing some kind of magic metaprogramming hackery that I don’t understand:
as.character.Surv <- function(x, ...) {
if (is.R()) class(x) <- NULL
else oldClass(x) <- NULL
type <- attr(x, 'type')
if (type=='right') {
temp <- x[,2]
temp <- ifelse(is.na(temp), "?", ifelse(temp==0, "+"," "))
paste(format(x[,1]), temp, sep='')
}
else if (type=='counting') {
temp <- x[,3]
temp <- ifelse(is.na(temp), "?", ifelse(temp==0, "+"," "))
paste('(', format(x[,1]), ',', format(x[,2]), temp,
']', sep='')
}
else if (type=='left') {
temp <- x[,2]
temp <- ifelse(is.na(temp), "?", ifelse(temp==0, "<"," "))
paste(temp, format(x[,1]), sep='')
}
else { #interval type
stat <- x[,3]
temp <- c("+", "", "-", "]")[stat+1]
temp2 <- ifelse(stat==3,
paste("[", format(x[,1]), ", ",format(x[,2]), sep=''),
format(x[,1]))
ifelse(is.na(stat), "NA", paste(temp2, temp, sep=''))
}
}
A method from Cynthiune, an MP3 player written in Objective-C for GNUStep, showing the old style retain/release/autorelease memory-management for NeXTSTep apps:
@implementation M3UUnarchiver : PlaylistUnarchiver
+ (NSArray *) _fileListFromLines: (NSArray *) arrayOfLines
inReferenceDirectory: (NSString *) directory
{
NSMutableArray *filelist;
NSEnumerator *lines;
NSString *currLine, *newString;
filelist = [NSMutableArray new];
[filelist autorelease];
lines = [arrayOfLines objectEnumerator];
currLine = [lines nextObject];
while (currLine)
{
if (![currLine hasPrefix: @"#"] && [currLine length])
{
newString = [NSString stringWithString: currLine];
if (![newString isAbsolutePath])
newString = [directory stringByAppendingPathComponent: newString];
newString = [newString stringByStandardizingPath];
[filelist addObject: newString];
}
currLine = [lines nextObject];
}
return filelist;
}
A C part of the SQLite test suite:
/*
** Usage: sqlite3_shared_cache_report
**
** Return a list of file that are shared and the number of
** references to each file.
*/
int sqlite3BtreeSharedCacheReport(
void * clientData,
Tcl_Interp *interp,
int objc,
Tcl_Obj *CONST objv[]
){
#ifndef SQLITE_OMIT_SHARED_CACHE
extern BtShared *sqlite3SharedCacheList;
BtShared *pBt;
Tcl_Obj *pRet = Tcl_NewObj();
for(pBt=GLOBAL(BtShared*,sqlite3SharedCacheList); pBt; pBt=pBt->pNext){
const char *zFile = sqlite3PagerFilename(pBt->pPager, 1);
Tcl_ListObjAppendElement(interp, pRet, Tcl_NewStringObj(zFile, -1));
Tcl_ListObjAppendElement(interp, pRet, Tcl_NewIntObj(pBt->nRef));
}
Tcl_SetObjResult(interp, pRet);
#endif
return TCL_OK;
}
A small part of the SQLite test suite, in Tcl:
# Convert a string containing EXPR, AGG, and BOOL into a string
# that contains nothing but X, Y, and Z.
#
proc extract_vars {a} {
regsub -all {EXPR} $a X a
regsub -all {AGG} $a Y a
regsub -all {BOOL} $a Z a
regsub -all {[^XYZ]} $a {} a
return $a
}
# Test all templates to make sure the number of EXPR, AGG, and BOOL
# expressions match.
#
foreach term [concat $aggexpr $intexpr $boolexpr] {
foreach {a b} $term break
if {[extract_vars $a]!=[extract_vars $b]} {
error "mismatch: $term"
}
}
From some dumb test code I wrote for a C random music generator:
int
main()
{
int ii, jj, kk;
score sc;
srand(getpid());
generate_score(&sc);
for (ii = 0; ii != 256; ii++) {
for (jj = 0; jj != 8192; jj++) {
unsigned char samp = 0;
for (kk = 0; kk < 6; kk++) {
int t = ii << (kk/2) | jj >> (13 - (kk/2));
int f = freq(next_note(t, &sc, kk));
samp += jj * f * (1 << kk) >> 10 & 32;
}
printf("%c", samp);
}
}
return 0;
}
More code from SQLite, this time in one of its memory allocators:
/*
** Unlink the chunk at mem5.aPool[i] from list it is currently
** on. It should be found on mem5.aiFreelist[iLogsize].
*/
static void memsys5Unlink(int i, int iLogsize){
int next, prev;
assert( i>=0 && i<mem5.nBlock );
assert( iLogsize>=0 && iLogsize<=LOGMAX );
assert( (mem5.aCtrl[i] & CTRL_LOGSIZE)==iLogsize );
next = MEM5LINK(i)->next;
prev = MEM5LINK(i)->prev;
if( prev<0 ){
mem5.aiFreelist[iLogsize] = next;
}else{
MEM5LINK(prev)->next = next;
}
if( next>=0 ){
MEM5LINK(next)->prev = prev;
}
}
A function in the Maybe monad from Brandon Moore’s Haskell ray tracer:
hit :: Vec -> Vec -> Sphere -> Maybe (Float, Vec, Vec)
hit x d (Sphere r center _ _) = do
let face = diff x center
a = dot d d
b = 2*dot d face
c = (dot face face) - (r*r)
disc = b^2 - 4*a*c
guard (b < 0) -- vector from center to camera should point opposite from ray
guard (disc >= 0)
let t = 2*c / (-b + sqrt disc)
intersection = x `add` scale t d
normal = normalize (diff intersection center)
xbounce = dot d normal
reflected = add d (scale (-2*xbounce) normal)
return (t, intersection, reflected)
From Lucene:
public class DateTools {
private final static TimeZone GMT = TimeZone.getTimeZone("GMT");
private static final SimpleDateFormat YEAR_FORMAT = new SimpleDateFormat("yyyy", Locale.US);
private static final SimpleDateFormat MONTH_FORMAT = new SimpleDateFormat("yyyyMM", Locale.US);
private static final SimpleDateFormat DAY_FORMAT = new SimpleDateFormat("yyyyMMdd", Locale.US);
private static final SimpleDateFormat HOUR_FORMAT = new SimpleDateFormat("yyyyMMddHH", Locale.US);
private static final SimpleDateFormat MINUTE_FORMAT = new SimpleDateFormat("yyyyMMddHHmm", Locale.US);
private static final SimpleDateFormat SECOND_FORMAT = new SimpleDateFormat("yyyyMMddHHmmss", Locale.US);
private static final SimpleDateFormat MILLISECOND_FORMAT = new SimpleDateFormat("yyyyMMddHHmmssSSS", Locale.US);
static {
// times need to be normalized so the value doesn't depend on the
// location the index is created/used:
YEAR_FORMAT.setTimeZone(GMT);
MONTH_FORMAT.setTimeZone(GMT);
DAY_FORMAT.setTimeZone(GMT);
HOUR_FORMAT.setTimeZone(GMT);
MINUTE_FORMAT.setTimeZone(GMT);
SECOND_FORMAT.setTimeZone(GMT);
MILLISECOND_FORMAT.setTimeZone(GMT);
}
private static final Calendar calInstance = Calendar.getInstance(GMT);
// cannot create, the class has static methods only
private DateTools() {}
// ...
}
From shlex in the Python standard library:
def read_token(self):
quoted = False
escapedstate = ' '
while True:
nextchar = self.instream.read(1)
if nextchar == '\n':
self.lineno = self.lineno + 1
if self.debug >= 3:
print "shlex: in state", repr(self.state), \
"I see character:", repr(nextchar)
if self.state is None:
self.token = '' # past end of file
break
elif self.state == ' ':
if not nextchar:
self.state = None # end of file
break
elif nextchar in self.whitespace:
if self.debug >= 2:
print "shlex: I see whitespace in whitespace state"
if self.token or (self.posix and quoted):
break # emit current token
else:
continue
elif nextchar in self.commenters:
self.instream.readline()
self.lineno = self.lineno + 1
elif self.posix and nextchar in self.escape:
escapedstate = 'a'
self.state = nextchar
elif nextchar in self.wordchars:
self.token = nextchar
self.state = 'a'
Some test code for a Visual Basic interface to mod_pubsub:
Module TestUtil
Public TU_OK As String = "OK"
Dim TU_VERBOSE_ALL As Boolean = False
Dim TU_TESTSUITE As String = ""
Dim TU_TESTCASE As String = ""
Dim TU_SERVER As String = ""
Dim TU_MISSINGSERVER As String = ""
Dim TU_SERVERNAME As String = ""
Dim tu_servers As New ArrayList()
Dim tu_count_subListener_OnUpdate As Integer = 0
Dim tu_count_subConnSH_OnConnStatus As Integer = 0
Dim tu_count_subReqSH_OnError As Integer = 0
Dim tu_count_subReqSH_OnSuccess As Integer = 0
Dim tu_count_pubConnSH_OnConnStatus As Integer = 0
Dim tu_count_pubReqSH_OnError As Integer = 0
Dim tu_count_pubReqSH_OnSuccess As Integer = 0
'/////////////////////////////////////////////////////////////////////////
'/// Public Events
'/////////////////////////////////////////////////////////////////////////
Public Class TU_SubListener
Inherits LibKNDotNet.IListener
Public Overrides Sub OnUpdate(ByVal msg As LibKNDotNet.Message)
TU_dumpMsg("TU_SubListener.OnUpdate", msg)
tu_count_subListener_OnUpdate = tu_count_subListener_OnUpdate + 1
End Sub
End Class
Public Sub TU_INIT_TESTCASE(ByVal testCaseName As String)
TU_TESTCASE = testCaseName
If TU_VERBOSE_ALL = False Then
Return
End If
Console.WriteLine()
Console.WriteLine("===============================================================================")
Console.WriteLine("=== Start test: " & TU_TESTSUITE & "." & testCaseName)
Console.WriteLine("===============================================================================")
End Sub
' lots more stuff here omitted
End Module
A PHP class from Laravel:
namespace App\Providers;
use Illuminate\Contracts\Auth\Access\Gate as GateContract;
use Illuminate\Foundation\Support\Providers\AuthServiceProvider as ServiceProvider;
class AuthServiceProvider extends ServiceProvider
{
/**
* The policy mappings for the application.
*
* @var array
*/
protected $policies = [
'App\Model' => 'App\Policies\ModelPolicy',
];
/**
* Register any application authentication / authorization services.
*
* @param \Illuminate\Contracts\Auth\Access\Gate $gate
* @return void
*/
public function boot(GateContract $gate)
{
$this->registerPolicies($gate);
//
}
}
A database schema migration in PHP from Laravel:
class CreateUsersTable extends Migration
{
/**
* Run the migrations.
*
* @return void
*/
public function up()
{
Schema::create('users', function (Blueprint $table) {
$table->increments('id');
$table->string('name');
$table->string('email')->unique();
$table->string('password');
$table->rememberToken();
$table->timestamps();
});
}
/**
* Reverse the migrations.
*
* @return void
*/
public function down()
{
Schema::drop('users');
}
}
Part of the Python 3.2 marshal module:
static void
w_object(PyObject *v, WFILE *p)
{
Py_ssize_t i, n;
p->depth++;
if (p->depth > MAX_MARSHAL_STACK_DEPTH) {
p->error = WFERR_NESTEDTOODEEP;
}
else if (v == NULL) {
w_byte(TYPE_NULL, p);
}
else if (v == Py_None) {
w_byte(TYPE_NONE, p);
}
else if (v == PyExc_StopIteration) {
w_byte(TYPE_STOPITER, p);
}
else if (v == Py_Ellipsis) {
w_byte(TYPE_ELLIPSIS, p);
}
else if (v == Py_False) {
w_byte(TYPE_FALSE, p);
}
else if (v == Py_True) {
w_byte(TYPE_TRUE, p);
}
else if (PyLong_CheckExact(v)) {
long x = PyLong_AsLong(v);
if ((x == -1) && PyErr_Occurred()) {
PyLongObject *ob = (PyLongObject *)v;
PyErr_Clear();
w_PyLong(ob, p);
}
else {
#if SIZEOF_LONG > 4
long y = Py_ARITHMETIC_RIGHT_SHIFT(long, x, 31);
if (y && y != -1) {
w_byte(TYPE_INT64, p);
w_long64(x, p);
}
else
#endif
{
w_byte(TYPE_INT, p);
w_long(x, p);
}
}
}
// other cases omitted
}
A three-page function from GNU cp:
static bool
make_dir_parents_private (char const *const_dir, size_t src_offset,
char const *verbose_fmt_string,
struct dir_attr **attr_list, bool *new_dst,
const struct cp_options *x)
{
struct stat stats;
char *dir; /* A copy of CONST_DIR we can change. */
char *src; /* Source name in DIR. */
char *dst_dir; /* Leading directory of DIR. */
size_t dirlen; /* Length of DIR. */
ASSIGN_STRDUPA (dir, const_dir);
src = dir + src_offset;
dirlen = dir_len (dir);
dst_dir = alloca (dirlen + 1);
memcpy (dst_dir, dir, dirlen);
dst_dir[dirlen] = '\0';
*attr_list = NULL;
if (stat (dst_dir, &stats) != 0)
{
/* A parent of CONST_DIR does not exist.
Make all missing intermediate directories. */
char *slash;
slash = src;
while (*slash == '/')
slash++;
while ((slash = strchr (slash, '/')))
{
struct dir_attr *new IF_LINT ( = NULL);
bool missing_dir;
*slash = '\0';
missing_dir = (stat (dir, &stats) != 0);
if (missing_dir || x->preserve_ownership || x->preserve_mode
|| x->preserve_timestamps)
{
/* Add this directory to the list of directories whose
modes might need fixing later. */
struct stat src_st;
int src_errno = (stat (src, &src_st) != 0
? errno
: S_ISDIR (src_st.st_mode)
? 0
: ENOTDIR);
if (src_errno)
{
error (0, src_errno, _("failed to get attributes of %s"),
quote (src));
return false;
}
new = xmalloc (sizeof *new);
new->st = src_st;
new->slash_offset = slash - dir;
new->restore_mode = false;
new->next = *attr_list;
*attr_list = new;
}
if (missing_dir)
{
mode_t src_mode;
mode_t omitted_permissions;
mode_t mkdir_mode;
/* This component does not exist. We must set
*new_dst and new->st.st_mode inside this loop because,
for example, in the command `cp --parents ../a/../b/c e_dir',
make_dir_parents_private creates only e_dir/../a if
./b already exists. */
*new_dst = true;
src_mode = new->st.st_mode;
/* If the ownership or special mode bits might change,
omit some permissions at first, so unauthorized users
cannot nip in before the file is ready. */
omitted_permissions = (src_mode
& (x->preserve_ownership
? S_IRWXG | S_IRWXO
: x->preserve_mode
? S_IWGRP | S_IWOTH
: 0));
/* POSIX says mkdir's behavior is implementation-defined when
(src_mode & ~S_IRWXUGO) != 0. However, common practice is
to ask mkdir to copy all the CHMOD_MODE_BITS, letting mkdir
decide what to do with S_ISUID | S_ISGID | S_ISVTX. */
mkdir_mode = src_mode & CHMOD_MODE_BITS & ~omitted_permissions;
if (mkdir (dir, mkdir_mode) != 0)
{
error (0, errno, _("cannot make directory %s"),
quote (dir));
return false;
}
else
{
if (verbose_fmt_string != NULL)
printf (verbose_fmt_string, src, dir);
}
/* We need search and write permissions to the new directory
for writing the directory's contents. Check if these
permissions are there. */
if (lstat (dir, &stats))
{
error (0, errno, _("failed to get attributes of %s"),
quote (dir));
return false;
}
if (! x->preserve_mode)
{
if (omitted_permissions & ~stats.st_mode)
omitted_permissions &= ~ cached_umask ();
if (omitted_permissions & ~stats.st_mode
|| (stats.st_mode & S_IRWXU) != S_IRWXU)
{
new->st.st_mode = stats.st_mode | omitted_permissions;
new->restore_mode = true;
}
}
if ((stats.st_mode & S_IRWXU) != S_IRWXU)
{
/* Make the new directory searchable and writable.
The original permissions will be restored later. */
if (lchmod (dir, stats.st_mode | S_IRWXU) != 0)
{
error (0, errno, _("setting permissions for %s"),
quote (dir));
return false;
}
}
}
else if (!S_ISDIR (stats.st_mode))
{
error (0, 0, _("%s exists but is not a directory"),
quote (dir));
return false;
}
else
*new_dst = false;
*slash++ = '/';
/* Avoid unnecessary calls to `stat' when given
file names containing multiple adjacent slashes. */
while (*slash == '/')
slash++;
}
}
/* We get here if the parent of DIR already exists. */
else if (!S_ISDIR (stats.st_mode))
{
error (0, 0, _("%s exists but is not a directory"), quote (dst_dir));
return false;
}
else
{
*new_dst = false;
}
return true;
}