mirror of
https://github.com/johnkerl/miller.git
synced 2026-07-25 17:04:01 +00:00
300 lines
13 KiB
C
300 lines
13 KiB
C
#include "lib/mlr_globals.h"
|
|
#include "lib/mlrutil.h"
|
|
#include "containers/lrec.h"
|
|
#include "containers/sllv.h"
|
|
#include "mapping/mappers.h"
|
|
#include "mapping/lrec_evaluators.h"
|
|
#include "dsls/mlr_dsl_wrapper.h"
|
|
#include "cli/argparse.h"
|
|
|
|
typedef struct _mapper_put_state_t {
|
|
ap_state_t* pargp;
|
|
sllv_t* pasts;
|
|
int num_evaluators;
|
|
int gate_invert;
|
|
int gate_bypass;
|
|
char** output_field_names;
|
|
lrec_evaluator_t** pevaluators;
|
|
} mapper_put_state_t;
|
|
|
|
static sllv_t* mapper_put_process(lrec_t* pinrec, context_t* pctx, void* pvstate);
|
|
static void mapper_put_free(mapper_t* pmapper);
|
|
static mapper_t* mapper_put_alloc(ap_state_t* pargp, sllv_t* pasts, int type_inferencing,
|
|
int gate_invert, int gate_bypass);
|
|
static void mapper_put_usage(FILE* o, char* argv0, char* verb);
|
|
static mapper_t* mapper_put_parse_cli(int* pargi, int argc, char** argv);
|
|
|
|
// ----------------------------------------------------------------
|
|
mapper_setup_t mapper_put_setup = {
|
|
.verb = "put",
|
|
.pusage_func = mapper_put_usage,
|
|
.pparse_func = mapper_put_parse_cli
|
|
};
|
|
|
|
// ----------------------------------------------------------------
|
|
static void mapper_put_usage(FILE* o, char* argv0, char* verb) {
|
|
fprintf(o, "Usage: %s %s [options] {expression}\n", argv0, verb);
|
|
fprintf(o, "Adds/updates specified field(s). Expressions are semicolon-separated and must\n");
|
|
fprintf(o, "either be assignments, or evaluate to boolean. Each expression is evaluated in\n");
|
|
fprintf(o, "turn from left to right. Assignment expressions are applied to the current\n");
|
|
fprintf(o, "record; once a boolean expression evaluates to false, the record is emitted\n");
|
|
fprintf(o, "with all changes up to that point and remaining expressions to the right are\n");
|
|
fprintf(o, "not evaluated.\n");
|
|
fprintf(o, "\n");
|
|
fprintf(o, "Options:\n");
|
|
fprintf(o, "-v: First prints the AST (abstract syntax tree) for the expression, which gives\n");
|
|
fprintf(o, " full transparency on the precedence and associativity rules of Miller's\n");
|
|
fprintf(o, " grammar.\n");
|
|
fprintf(o, "-S: Keeps field values, or literals in the expression, as strings with no type \n");
|
|
fprintf(o, " inference to int or float.\n");
|
|
fprintf(o, "-F: Keeps field values, or literals in the expression, as strings or floats\n");
|
|
fprintf(o, " with no inference to int.\n");
|
|
fprintf(o, "-x: Negates boolean expressions. Has no effect on assignment expressions.\n");
|
|
fprintf(o, "-t: Continue evaluating remaining expressions after boolean expressions, whether\n");
|
|
fprintf(o, " or not they evaluate to true. Has no effect on assignment expressions.\n");
|
|
fprintf(o, "\n");
|
|
fprintf(o, "Please use a dollar sign for field names and double-quotes for string\n");
|
|
fprintf(o, "literals. If field names have special characters such as \".\" then you might\n");
|
|
fprintf(o, "use braces, e.g. '${field.name}'. Miller built-in variables are\n");
|
|
fprintf(o, "NF NR FNR FILENUM FILENAME PI E.\n");
|
|
fprintf(o, "\n");
|
|
fprintf(o, "Examples:\n");
|
|
fprintf(o, " Assignment only:\n");
|
|
fprintf(o, " %s %s '$y = log10($x); $z = sqrt($y)'\n", argv0, verb);
|
|
fprintf(o, " %s %s '$filename = FILENAME'\n", argv0, verb);
|
|
fprintf(o, " %s %s '$colored_shape = $color . \"_\" . $shape'\n", argv0, verb);
|
|
fprintf(o, " %s %s '$y = cos($theta); $z = atan2($y, $x)'\n", argv0, verb);
|
|
fprintf(o, " %s %s '$name = sub($name, \"http.*com\"i, \"\")'\n", argv0, verb);
|
|
fprintf(o, " Mixed assignment/boolean:\n");
|
|
fprintf(o, " %s %s '$x > 0.0; $y = log10($x); $z = sqrt($y)'\n", argv0, verb);
|
|
fprintf(o, " %s %s '$y = log10($x); 1.1 < $y && $y < 7.0; $z = sqrt($y)'\n", argv0, verb);
|
|
fprintf(o, "\n");
|
|
fprintf(o, "Please see http://johnkerl.org/miller/doc/reference.html for more information\n");
|
|
fprintf(o, "including function list. Or \"%s -f\".\n", argv0);
|
|
}
|
|
|
|
// ----------------------------------------------------------------
|
|
static mapper_t* mapper_put_parse_cli(int* pargi, int argc, char** argv) {
|
|
char* verb = argv[(*pargi)++];
|
|
char* mlr_dsl_expression = NULL;
|
|
int gate_invert = FALSE;
|
|
int gate_bypass = FALSE;
|
|
int type_inferencing = TYPE_INFER_STRING_FLOAT_INT;
|
|
int print_asts = FALSE;
|
|
|
|
ap_state_t* pstate = ap_alloc();
|
|
ap_define_true_flag(pstate, "-v", &print_asts);
|
|
ap_define_int_value_flag(pstate, "-S", TYPE_INFER_STRING_ONLY, &type_inferencing);
|
|
ap_define_int_value_flag(pstate, "-F", TYPE_INFER_STRING_FLOAT, &type_inferencing);
|
|
ap_define_true_flag(pstate, "-x", &gate_invert);
|
|
ap_define_true_flag(pstate, "-t", &gate_bypass);
|
|
|
|
if (!ap_parse(pstate, verb, pargi, argc, argv)) {
|
|
mapper_put_usage(stderr, argv[0], verb);
|
|
return NULL;
|
|
}
|
|
|
|
if ((argc - *pargi) < 1) {
|
|
mapper_put_usage(stderr, argv[0], verb);
|
|
return NULL;
|
|
}
|
|
mlr_dsl_expression = argv[(*pargi)++];
|
|
|
|
// Linked list of mlr_dsl_ast_node_t*.
|
|
sllv_t* pasts = mlr_dsl_parse(mlr_dsl_expression);
|
|
if (pasts == NULL) {
|
|
mapper_put_usage(stderr, argv[0], verb);
|
|
return NULL;
|
|
}
|
|
|
|
// For just dev-testing the parser, you can do
|
|
// mlr put -v 'expression goes here' /dev/null
|
|
if (print_asts) {
|
|
for (sllve_t* pe = pasts->phead; pe != NULL; pe = pe->pnext)
|
|
mlr_dsl_ast_node_print(pe->pvvalue);
|
|
}
|
|
|
|
return mapper_put_alloc(pstate, pasts, type_inferencing, gate_invert, gate_bypass);
|
|
}
|
|
|
|
// ----------------------------------------------------------------
|
|
static mapper_t* mapper_put_alloc(ap_state_t* pargp, sllv_t* pasts, int type_inferencing,
|
|
int gate_invert, int gate_bypass)
|
|
{
|
|
mapper_put_state_t* pstate = mlr_malloc_or_die(sizeof(mapper_put_state_t));
|
|
pstate->pargp = pargp;
|
|
pstate->pasts = pasts;
|
|
pstate->num_evaluators = pasts->length;
|
|
pstate->output_field_names = mlr_malloc_or_die(pasts->length * sizeof(char*));
|
|
pstate->gate_invert = gate_invert;
|
|
pstate->gate_bypass = gate_bypass;
|
|
pstate->pevaluators = mlr_malloc_or_die(pasts->length * sizeof(lrec_evaluator_t*));
|
|
|
|
int i = 0;
|
|
for (sllve_t* pe = pasts->phead; pe != NULL; pe = pe->pnext, i++) {
|
|
mlr_dsl_ast_node_t* past = pe->pvvalue;
|
|
|
|
if ((past->type == MLR_DSL_AST_NODE_TYPE_OPERATOR) && streq(past->text, "=")) {
|
|
// Assignment statement
|
|
if ((past->pchildren == NULL) || (past->pchildren->length != 2)) {
|
|
fprintf(stderr, "%s: internal coding error detected in file %s at line %d.\n",
|
|
MLR_GLOBALS.argv0, __FILE__, __LINE__);
|
|
exit(1);
|
|
}
|
|
|
|
mlr_dsl_ast_node_t* pleft = past->pchildren->phead->pvvalue;
|
|
mlr_dsl_ast_node_t* pright = past->pchildren->phead->pnext->pvvalue;
|
|
|
|
if (pleft->type != MLR_DSL_AST_NODE_TYPE_FIELD_NAME) {
|
|
fprintf(stderr, "%s: internal coding error detected in file %s at line %d.\n",
|
|
MLR_GLOBALS.argv0, __FILE__, __LINE__);
|
|
exit(1);
|
|
} else if (pleft->pchildren != NULL) {
|
|
fprintf(stderr, "%s: coding error detected in file %s at line %d.\n",
|
|
MLR_GLOBALS.argv0, __FILE__, __LINE__);
|
|
exit(1);
|
|
}
|
|
|
|
char* output_field_name = pleft->text;
|
|
lrec_evaluator_t* pevaluator = lrec_evaluator_alloc_from_ast(pright, type_inferencing);
|
|
pstate->pevaluators[i] = pevaluator;
|
|
pstate->output_field_names[i] = output_field_name;
|
|
|
|
} else {
|
|
// Gate statement
|
|
lrec_evaluator_t* pevaluator = lrec_evaluator_alloc_from_ast(past, type_inferencing);
|
|
pstate->pevaluators[i] = pevaluator;
|
|
pstate->output_field_names[i] = NULL;
|
|
}
|
|
}
|
|
|
|
mapper_t* pmapper = mlr_malloc_or_die(sizeof(mapper_t));
|
|
|
|
pmapper->pvstate = (void*)pstate;
|
|
pmapper->pprocess_func = mapper_put_process;
|
|
pmapper->pfree_func = mapper_put_free;
|
|
|
|
return pmapper;
|
|
}
|
|
|
|
static void mapper_put_free(mapper_t* pmapper) {
|
|
mapper_put_state_t* pstate = pmapper->pvstate;
|
|
free(pstate->output_field_names);
|
|
|
|
for (int i = 0; i < pstate->num_evaluators; i++) {
|
|
lrec_evaluator_t* pevaluator = pstate->pevaluators[i];
|
|
pevaluator->pfree_func(pevaluator);
|
|
}
|
|
free(pstate->pevaluators);
|
|
|
|
for (sllve_t* pe = pstate->pasts->phead; pe != NULL; pe = pe->pnext) {
|
|
mlr_dsl_ast_node_t* past = pe->pvvalue;
|
|
mlr_dsl_ast_node_free(past);
|
|
}
|
|
sllv_free(pstate->pasts);
|
|
|
|
ap_free(pstate->pargp);
|
|
|
|
free(pstate);
|
|
free(pmapper);
|
|
}
|
|
|
|
// ----------------------------------------------------------------
|
|
// The typed-overlay holds intermediate values such as in
|
|
//
|
|
// echo x=1 | mlr put '$y = string($x); $z = $y . $y'
|
|
//
|
|
// because otherwise
|
|
// * lrecs store insertion-ordered maps of string to string (since this is ultimately file I/O);
|
|
// * types are inferred at entry to put;
|
|
// * x=1 would be inferred to int; string($x) would be string; written back to the
|
|
// lrec, y would be "1" which would be re-inferred to int.
|
|
//
|
|
// So the typed overlay allows us to remember that y is string "1" not integer 1.
|
|
//
|
|
// But this raises the question: why stop here? Why not have lrecs be insertion-ordered maps from
|
|
// string to mlrval? Then we could preserve types for the duration of each lrec, not just for
|
|
// the duration of the put operation. Reasons:
|
|
// * The compare_lexically operation would suffer a performance regression;
|
|
// * Worse, all lhmslv group-by operations (used by many Miller verbs) would likewise suffer a performance regression.
|
|
//
|
|
// ----------------------------------------------------------------
|
|
// The regex-capture string-array holds copies of regex matches, e.g. in
|
|
//
|
|
// echo name=abc_def | mlr put '$name =~ "(.*)_(.*)"; $left = "\1"; $right = "\2"'
|
|
//
|
|
// produces a record with left=abc and right=def.
|
|
//
|
|
// There is an important trick here with the length of the string-array:
|
|
// * It is allocated here with length 0.
|
|
// * It is passed by reference to the lrec-evaluator tree. In particular, the matches and does-not-match functions
|
|
// (which implement the =~ and !=~ operators) resize it and populate it.
|
|
// * For simplicity, it is a 1-up array: so \1, \2, \3 are at array indices 1, 2, 3.
|
|
// * If the matches/does-not-match functions are entered, even with no matches, the regex-captures string-array
|
|
// will be resized to have length at least 1: length 1 for 0 matches, length 2 for 1 match, etc. since
|
|
// the array is indexed 1-up.
|
|
// * When the lrec-evaluator's from-literal function is invoked, the interpolate_regex_captures function can quickly
|
|
// check to see if the regex-captures array has length 0 and thereby know that a time-consuming scan for \1, \2, \3,
|
|
// etc. does not need to be done.
|
|
|
|
static sllv_t* mapper_put_process(lrec_t* pinrec, context_t* pctx, void* pvstate) {
|
|
if (pinrec == NULL) // End of input stream
|
|
return sllv_single(NULL);
|
|
|
|
mapper_put_state_t* pstate = (mapper_put_state_t*)pvstate;
|
|
lhmsv_t* ptyped_overlay = lhmsv_alloc();
|
|
string_array_t* pregex_captures = string_array_alloc(0);
|
|
|
|
// Do the evaluations, writing typed mlrval output to the typed overlay rather than into the lrec (which holds only
|
|
// string values).
|
|
for (int i = 0; i < pstate->num_evaluators; i++) {
|
|
lrec_evaluator_t* pevaluator = pstate->pevaluators[i];
|
|
char* output_field_name = pstate->output_field_names[i];
|
|
if (output_field_name != NULL) {
|
|
|
|
// Assignment statement
|
|
mv_t val = pevaluator->pprocess_func(pinrec, ptyped_overlay, pregex_captures, pctx, pevaluator->pvstate);
|
|
mv_t* pval = mlr_malloc_or_die(sizeof(mv_t));
|
|
*pval = val;
|
|
lhmsv_put(ptyped_overlay, output_field_name, pval, NO_FREE);
|
|
// The lrec_evaluator reads the overlay in preference to the lrec. E.g. if the input had
|
|
// "x"=>"abc","y"=>"def" but the previous pass through this loop set "y"=>7.4 and "z"=>"ghi" then an
|
|
// expression right-hand side referring to $y would get the floating-point value 7.4. So we don't need to do
|
|
// lrec_put here, and moreover should not for two reasons: (1) there is a performance hit of doing throwaway
|
|
// number-to-string formatting -- it's better to do it once at the end; (2) having the string values doubly
|
|
// owned by the typed overlay and the lrec would result in double frees, or awkward bookkeeping. However,
|
|
// the NR variable evaluator reads prec->field_count, so we need to put something here. And putting
|
|
// something statically allocated minimizes copying/freeing.
|
|
lrec_put(pinrec, output_field_name, "bug", NO_FREE);
|
|
|
|
} else {
|
|
|
|
// Gate statement
|
|
mv_t val = pevaluator->pprocess_func(pinrec, ptyped_overlay, pregex_captures, pctx, pevaluator->pvstate);
|
|
if (val.type == MT_NULL)
|
|
break;
|
|
mv_set_boolean_strict(&val);
|
|
if (!((val.u.boolv ^ pstate->gate_invert) | pstate->gate_bypass))
|
|
break;
|
|
|
|
}
|
|
}
|
|
|
|
// Write the output fields from the typed overlay back to the lrec.
|
|
for (lhmsve_t* pe = ptyped_overlay->phead; pe != NULL; pe = pe->pnext) {
|
|
char* output_field_name = pe->key;
|
|
mv_t* pval = pe->pvvalue;
|
|
if (pval->type == MT_STRING) {
|
|
// Ownership transfer from mv_t to lrec.
|
|
lrec_put(pinrec, output_field_name, pval->u.strv, pval->free_flags);
|
|
} else {
|
|
char free_flags = NO_FREE;
|
|
char* string = mv_format_val(pval, &free_flags);
|
|
lrec_put(pinrec, output_field_name, string, free_flags);
|
|
}
|
|
free(pval);
|
|
}
|
|
lhmsv_free(ptyped_overlay);
|
|
|
|
return sllv_single(pinrec);
|
|
}
|