mirror of
https://github.com/Unidata/netcdf-c.git
synced 2024-11-27 07:30:33 +08:00
521 lines
13 KiB
C
521 lines
13 KiB
C
/*********************************************************************
|
|
* Copyright 2018, UCAR/Unidata
|
|
* See netcdf/COPYRIGHT file for copying and redistribution conditions.
|
|
*********************************************************************/
|
|
|
|
#include <stdarg.h>
|
|
#include <assert.h>
|
|
#include <stddef.h>
|
|
|
|
#include "d4includes.h"
|
|
|
|
/*
|
|
The primary purpose of this code is to provide node and data walkers
|
|
to do a variety of things.
|
|
|
|
* (topo)
|
|
a. topologically sort the set of allnodes in postfix order.
|
|
|
|
* (delimit)
|
|
a. Delimit the top-level vars in the dap4data and also compute checksum
|
|
|
|
* (move)
|
|
a. Walk the (toplevel) var's data to get to the count'th instance.
|
|
|
|
*/
|
|
|
|
/**************************************************/
|
|
|
|
#define COUNTSIZE 8
|
|
|
|
/**************************************************/
|
|
/* Forward */
|
|
|
|
static int delimitAtomicVar(NCD4meta*, NCD4node* var, NCD4offset* offset);
|
|
static int delimitOpaqueVar(NCD4meta*, NCD4node* var, NCD4offset* offset);
|
|
static int delimitSeq(NCD4meta*, NCD4node* var, NCD4offset* offset);
|
|
static int delimitSeqArray(NCD4meta*, NCD4node* var, NCD4offset* offset);
|
|
static int delimitStruct(NCD4meta*, NCD4node* var, NCD4offset* offset);
|
|
static int delimitStructArray(NCD4meta*, NCD4node* var, NCD4offset* offset);
|
|
static int skipInstance(NCD4meta*, NCD4node* type, NCD4offset* offset);
|
|
static int skipAtomicInstance(NCD4meta*, NCD4node* type, NCD4offset* offset);
|
|
static int skipStructInstance(NCD4meta*, NCD4node* type, NCD4offset* offset);
|
|
static int skipSeqInstance(NCD4meta*, NCD4node* vlentype, NCD4offset* offset);
|
|
|
|
static void walk(NCD4node* node, NClist* sorted);
|
|
|
|
#ifdef D4DEBUGDATA
|
|
static void
|
|
ADDNODE(NClist* list, NCD4node* node)
|
|
{
|
|
fprintf(stderr,"addnode: %s: %s %s (%p)\n",
|
|
node->name,
|
|
NCD4_sortname(node->sort),
|
|
NCD4_subsortname(node->subsort),
|
|
node);
|
|
fflush(stderr);
|
|
nclistpush(list,node);
|
|
}
|
|
#else
|
|
#define ADDNODE(list,node) nclistpush((list),(node))
|
|
#endif
|
|
|
|
|
|
|
|
/**************************************************/
|
|
/* Topo sort in postfix order */
|
|
|
|
int
|
|
NCD4_toposort(NCD4meta* compiler)
|
|
{
|
|
int i, ret=NC_NOERR;
|
|
size_t len = nclistlength(compiler->allnodes);
|
|
NCD4node** list = (NCD4node**)nclistcontents(compiler->allnodes);
|
|
NCD4node** p;
|
|
NClist* sorted = nclistnew();
|
|
nclistsetalloc(sorted,len);
|
|
for(i=0,p=list;i<len;p++,i++) {
|
|
NCD4node* node = *p;
|
|
switch (node->sort) { /* Collect things known to have no dependencies */
|
|
case NCD4_DIM:
|
|
node->visited = 1;
|
|
ADDNODE(sorted,node);
|
|
break;
|
|
case NCD4_TYPE:
|
|
if(node->subsort <= NC_MAX_ATOMIC_TYPE || node->subsort == NC_OPAQUE) {
|
|
node->visited = 1;
|
|
ADDNODE(sorted,node);
|
|
} else
|
|
node->visited = 0;
|
|
break;
|
|
default:
|
|
node->visited = 0;
|
|
}
|
|
}
|
|
walk(compiler->root,sorted);
|
|
/* Last step is to add in any remaining unvisited nodes, but report them */
|
|
for(i=0,p=list;i<len;p++,i++) {
|
|
NCD4node* node = *p;
|
|
if(node->visited) continue;
|
|
#ifdef D4DEBUGDATA
|
|
fprintf(stderr,"unvisited node: %s\n",node->name); fflush(stderr);
|
|
#endif
|
|
node->visited = 1;
|
|
ADDNODE(sorted,node);
|
|
}
|
|
nclistfree(compiler->allnodes);
|
|
compiler->allnodes = sorted;
|
|
#ifdef D4DEBUGDATA
|
|
{int i;
|
|
for(i=0;i<nclistlength(sorted);i++)
|
|
fprintf(stderr,"sorted: %s\n",((NCD4node*)nclistget(sorted,i))->name);
|
|
fflush(stderr);
|
|
}
|
|
#endif
|
|
return THROW(ret);
|
|
}
|
|
|
|
/*
|
|
Do depth first search
|
|
*/
|
|
static void
|
|
walk(NCD4node* node, NClist* sorted)
|
|
{
|
|
size_t i;
|
|
|
|
if(node->visited) return;
|
|
node->visited = 1;
|
|
|
|
switch (node->sort) {
|
|
case NCD4_GROUP: /* depends on its elements and attributes and subgroups */
|
|
for(i=0;i<nclistlength(node->group.elements);i++) {
|
|
NCD4node* elem = (NCD4node*)nclistget(node->group.elements,i);
|
|
walk(elem,sorted);
|
|
}
|
|
break;
|
|
case NCD4_TYPE: /* Need to discriminate on the subsort */
|
|
switch (node->subsort) {
|
|
case NC_SEQ:
|
|
/* Depends on its basetype */
|
|
walk(node->basetype,sorted);
|
|
break;
|
|
case NC_STRUCT: /* Depends on its fields */
|
|
for(i=0;i<nclistlength(node->vars);i++) {
|
|
NCD4node* f = (NCD4node*)nclistget(node->vars,i);
|
|
walk(f,sorted);
|
|
}
|
|
break;
|
|
case NC_ENUM: /* Depends on its basetype, but since that is atomic, we can ignore */
|
|
/* fall thru */
|
|
default: /* Atomic or opaque, so already marked */
|
|
break;
|
|
}
|
|
break;
|
|
|
|
case NCD4_VAR: /* Depends on: dimensions and basetype and maps */
|
|
for(i=0;i<nclistlength(node->dims);i++) {
|
|
NCD4node* d = (NCD4node*)nclistget(node->dims,i);
|
|
walk(d,sorted);
|
|
}
|
|
walk(node->basetype,sorted);
|
|
for(i=0;i<nclistlength(node->maps);i++) {
|
|
NCD4node* m = (NCD4node*)nclistget(node->maps,i);
|
|
walk(m,sorted);
|
|
}
|
|
break;
|
|
|
|
case NCD4_ATTR: /* Depends on its base type */
|
|
walk(node->basetype,sorted);
|
|
break;
|
|
|
|
case NCD4_ATTRSET: /* Depends on its contained attributes, but handled after switch */
|
|
/* fall thru */
|
|
default: /* depends on nothing else */
|
|
break;
|
|
}
|
|
|
|
/* Do Attributes last */
|
|
for(i=0;i<nclistlength(node->attributes);i++) {
|
|
NCD4node* a = (NCD4node*)nclistget(node->attributes,i);
|
|
walk(a,sorted);
|
|
}
|
|
ADDNODE(sorted,node);
|
|
}
|
|
|
|
|
|
/**************************************************/
|
|
/* Mark the offset and length of each var/field
|
|
inside the raw dapdata.
|
|
Assumes it is called before byte swapping, so we
|
|
need to do swapping of counts and the final remote checksum.
|
|
*/
|
|
|
|
int
|
|
NCD4_delimit(NCD4meta* compiler, NCD4node* topvar, NCD4offset* offset, int inferredchecksumming)
|
|
{
|
|
int ret = NC_NOERR;
|
|
NCD4mark mark = 0;
|
|
|
|
ASSERT((ISTOPLEVEL(topvar)));
|
|
MARK(offset,mark);
|
|
if(topvar->sort == NCD4_VAR) {
|
|
switch (topvar->subsort) {
|
|
case NC_STRUCT:
|
|
if((ret=delimitStructArray(compiler,topvar,offset))) goto done;
|
|
break;
|
|
case NC_SEQ:
|
|
if((ret=delimitSeqArray(compiler,topvar,offset))) goto done;
|
|
break;
|
|
default:
|
|
if((ret=delimitAtomicVar(compiler,topvar,offset))) goto done;
|
|
break;
|
|
}
|
|
}
|
|
/* Track the variable size (in the dap4 data) but do not include any checksum */
|
|
topvar->data.dap4data.memory = mark;
|
|
topvar->data.dap4data.size = OFFSETSIZE(offset,mark);
|
|
/* extract the dap4 data checksum, if present */
|
|
if(inferredchecksumming) {
|
|
union ATOMICS csum;
|
|
TRANSFER(csum.u8,offset,CHECKSUMSIZE);
|
|
topvar->data.remotechecksum = csum.u32[0];
|
|
if(compiler->swap) swapinline32(&topvar->data.remotechecksum);
|
|
INCR(offset,CHECKSUMSIZE);
|
|
}
|
|
done:
|
|
return THROW(ret);
|
|
}
|
|
|
|
/* Includes opaque and enum */
|
|
static int
|
|
delimitAtomicVar(NCD4meta* compiler, NCD4node* var, NCD4offset* offset)
|
|
{
|
|
int ret = NC_NOERR;
|
|
d4size_t i;
|
|
d4size_t dimproduct;
|
|
nc_type tid;
|
|
NCD4node* basetype;
|
|
NCD4node* truetype;
|
|
|
|
assert(var->sort == NCD4_VAR);
|
|
dimproduct = NCD4_dimproduct(var);
|
|
basetype = var->basetype;
|
|
if(basetype->subsort == NC_OPAQUE)
|
|
return delimitOpaqueVar(compiler,var,offset);
|
|
|
|
truetype = basetype;
|
|
if(truetype->subsort == NC_ENUM)
|
|
truetype = basetype->basetype;
|
|
|
|
tid = truetype->subsort;
|
|
size_t typesize = NCD4_typesize(tid);
|
|
if(tid != NC_STRING) {
|
|
INCR(offset,(typesize*dimproduct));
|
|
} else if(tid == NC_STRING) { /* walk the counts */
|
|
unsigned long long count;
|
|
for(i=0;i<dimproduct;i++) {
|
|
/* Get string count */
|
|
count = GETCOUNTER(offset);
|
|
SKIPCOUNTER(offset);
|
|
if(compiler->swap) swapinline64(&count);
|
|
/* skip count bytes */
|
|
INCR(offset,count);
|
|
}
|
|
}
|
|
return THROW(ret);
|
|
}
|
|
|
|
static int
|
|
delimitOpaqueVar(NCD4meta* compiler, NCD4node* var, NCD4offset* offset)
|
|
{
|
|
int ret = NC_NOERR;
|
|
d4size_t i;
|
|
unsigned long long count;
|
|
d4size_t dimproduct = NCD4_dimproduct(var);
|
|
|
|
for(i=0;i<dimproduct;i++) {
|
|
/* Walk the instances */
|
|
count = GETCOUNTER(offset);
|
|
SKIPCOUNTER(offset);
|
|
if(compiler->swap) swapinline64(&count);
|
|
INCR(offset,count);
|
|
}
|
|
return THROW(ret);
|
|
}
|
|
|
|
static int
|
|
delimitStructArray(NCD4meta* compiler, NCD4node* varortype, NCD4offset* offset)
|
|
{
|
|
int ret = NC_NOERR;
|
|
d4size_t i;
|
|
d4size_t dimproduct;
|
|
NCD4node* type;
|
|
|
|
if(varortype->sort == NCD4_VAR) {
|
|
dimproduct = NCD4_dimproduct(varortype);
|
|
type = varortype->basetype;
|
|
} else {
|
|
dimproduct = 1;
|
|
type = varortype;
|
|
}
|
|
|
|
for(i=0;i<dimproduct;i++) {
|
|
if((ret=delimitStruct(compiler,type,offset))) goto done;
|
|
}
|
|
done:
|
|
return THROW(ret);
|
|
}
|
|
|
|
static int
|
|
delimitStruct(NCD4meta* compiler, NCD4node* basetype, NCD4offset* offset)
|
|
{
|
|
int ret = NC_NOERR;
|
|
size_t i;
|
|
|
|
/* The fields are associated with the basetype struct */
|
|
for(i=0;i<nclistlength(basetype->vars);i++) {
|
|
NCD4node* field = (NCD4node*)nclistget(basetype->vars,i);
|
|
switch (field->subsort) {
|
|
default:
|
|
if((ret=delimitAtomicVar(compiler,field,offset))) goto done;
|
|
break;
|
|
case NC_STRUCT: /* recurse */
|
|
if((ret=delimitStructArray(compiler,field,offset))) goto done;
|
|
break;
|
|
case NC_SEQ:
|
|
if((ret=delimitSeqArray(compiler,field,offset))) goto done;
|
|
break;
|
|
}
|
|
}
|
|
done:
|
|
return THROW(ret);
|
|
}
|
|
|
|
static int
|
|
delimitSeqArray(NCD4meta* compiler, NCD4node* varortype, NCD4offset* offset)
|
|
{
|
|
int ret = NC_NOERR;
|
|
d4size_t i;
|
|
d4size_t dimproduct;
|
|
NCD4node* type;
|
|
|
|
if(varortype->sort == NCD4_VAR) {
|
|
dimproduct = NCD4_dimproduct(varortype);
|
|
type = varortype->basetype;
|
|
} else {
|
|
dimproduct = 1;
|
|
type = varortype;
|
|
}
|
|
|
|
for(i=0;i<dimproduct;i++) {
|
|
if((ret=delimitSeq(compiler,type,offset))) goto done;
|
|
}
|
|
|
|
done:
|
|
return THROW(ret);
|
|
}
|
|
|
|
static int
|
|
delimitSeq(NCD4meta* compiler, NCD4node* vlentype, NCD4offset* offset)
|
|
{
|
|
int ret = NC_NOERR;
|
|
int i;
|
|
d4size_t recordcount;
|
|
NCD4node* recordtype;
|
|
|
|
/* The true type of the record is the basetype of the vlen,
|
|
where the vlen type is the basetype of the var
|
|
*/
|
|
assert(vlentype->subsort == NC_VLEN);
|
|
recordtype = vlentype->basetype;
|
|
|
|
/* Get he record count */
|
|
recordcount = GETCOUNTER(offset);
|
|
SKIPCOUNTER(offset);
|
|
if(compiler->swap) swapinline64(&recordcount);
|
|
|
|
for(i=0;i<recordcount;i++) {
|
|
switch (recordtype->subsort) {
|
|
case NC_STRUCT:
|
|
if((ret=delimitStructArray(compiler,recordtype,offset))) goto done;
|
|
break;
|
|
case NC_SEQ:
|
|
if((ret=delimitSeqArray(compiler,recordtype,offset))) goto done;
|
|
break;
|
|
default:
|
|
if((ret=delimitAtomicVar(compiler,recordtype,offset))) goto done;
|
|
break;
|
|
}
|
|
}
|
|
|
|
done:
|
|
return THROW(ret);
|
|
}
|
|
|
|
/**************************************************/
|
|
/*
|
|
Walk the (toplevel) var's data to get to the count'th instance.
|
|
For efficiency, it can be supplied with a previous case.
|
|
Assumes it is called after byte swapping and offsetting.
|
|
Assumes that var is not fixed size.
|
|
*/
|
|
|
|
int
|
|
NCD4_moveto(NCD4meta* compiler, NCD4node* var, d4size_t count, NCD4offset* offset)
|
|
{
|
|
int ret = NC_NOERR;
|
|
d4size_t startcount = 0;
|
|
NCD4node* basetype = NULL;
|
|
|
|
ASSERT((ISTOPLEVEL(var)));
|
|
|
|
startcount = 0;
|
|
basetype = var->basetype;
|
|
for(;startcount < count;startcount++) {
|
|
if((ret=skipInstance(compiler,basetype,offset)))
|
|
goto done;
|
|
}
|
|
|
|
done:
|
|
return THROW(ret);
|
|
}
|
|
|
|
static int
|
|
skipInstance(NCD4meta* compiler, NCD4node* type, NCD4offset* offset)
|
|
{
|
|
int ret = NC_NOERR;
|
|
|
|
switch (type->subsort) {
|
|
case NC_STRUCT:
|
|
if((ret=skipStructInstance(compiler,type,offset))) goto done;
|
|
break;
|
|
case NC_SEQ:
|
|
if((ret=skipSeqInstance(compiler,type,offset))) goto done;
|
|
break;
|
|
default:
|
|
if((ret=skipAtomicInstance(compiler,type,offset))) goto done;
|
|
break;
|
|
}
|
|
|
|
done:
|
|
return THROW(ret);
|
|
}
|
|
|
|
/* Includes opaque and enum */
|
|
static int
|
|
skipAtomicInstance(NCD4meta* compiler, NCD4node* type, NCD4offset* offset)
|
|
{
|
|
int ret = NC_NOERR;
|
|
d4size_t count;
|
|
size_t typesize;
|
|
|
|
switch (type->subsort) {
|
|
default: /* fixed size atomic type */
|
|
typesize = NCD4_typesize(type->meta.id);
|
|
INCR(offset,typesize);
|
|
break;
|
|
case NC_STRING:
|
|
/* Get string count */
|
|
count = GETCOUNTER(offset);
|
|
SKIPCOUNTER(offset);
|
|
/* skip count bytes */
|
|
INCR(offset,count);
|
|
break;
|
|
case NC_OPAQUE:
|
|
/* get count */
|
|
count = GETCOUNTER(offset);
|
|
SKIPCOUNTER(offset);
|
|
INCR(offset,count);
|
|
break;
|
|
case NC_ENUM:
|
|
return skipAtomicInstance(compiler,type->basetype,offset);
|
|
}
|
|
return THROW(ret);
|
|
}
|
|
|
|
static int
|
|
skipStructInstance(NCD4meta* compiler, NCD4node* type, NCD4offset* offset)
|
|
{
|
|
int ret = NC_NOERR;
|
|
d4size_t i,j;
|
|
|
|
/* Skip each field */
|
|
for(i=0;i<nclistlength(type->vars);i++) {
|
|
NCD4node* field = (NCD4node*)nclistget(type->vars,i);
|
|
NCD4node* ftype = field->basetype;
|
|
d4size_t dimproduct = NCD4_dimproduct(field);
|
|
for(j=0;j<dimproduct;j++) {
|
|
if((ret=skipInstance(compiler,ftype,offset)))
|
|
goto done;
|
|
}
|
|
}
|
|
done:
|
|
return THROW(ret);
|
|
}
|
|
|
|
static int
|
|
skipSeqInstance(NCD4meta* compiler, NCD4node* vlentype, NCD4offset* offset)
|
|
{
|
|
int ret = NC_NOERR;
|
|
d4size_t i;
|
|
NCD4node* structtype;
|
|
d4size_t recordcount;
|
|
|
|
structtype = vlentype->basetype;
|
|
ASSERT((structtype->subsort == NC_STRUCT));
|
|
|
|
/* Get record count */
|
|
recordcount = GETCOUNTER(offset);
|
|
SKIPCOUNTER(offset);
|
|
|
|
for(i=0;i<recordcount;i++) {
|
|
/* Skip a record instance */
|
|
if((ret=skipStructInstance(compiler,structtype,offset)))
|
|
goto done;
|
|
}
|
|
|
|
done:
|
|
return THROW(ret);
|
|
}
|
|
|