ply/ply.c

#include <assert.h>
#include <errno.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>

#include "func.h"
#include "node.h"
#include "ply.h"
#include "sym.h"
#include "type.h"

struct providers {
	struct provider **ps;
	size_t len;
} providers;

#define providers_foreach(_ps, _p) \
	for((_p) = (_ps)->ps; (_p) < &(_ps)->ps[(_ps)->len]; (_p)++)

struct provider *provider_get(const char *name)
{
	struct provider **p;

	providers_foreach(&providers, p) {
		if (strstr((*p)->name, name) == (*p)->name)
			return *p;
	}

	return NULL;
}

void provider_register(struct provider *p)
{
	assert(p);
	assert(p->probe);
	assert(p->sym_alloc);

	providers.ps = realloc(providers.ps,
				 ++providers.len * sizeof(*providers.ps));

	providers.ps[providers.len - 1] = p;
}

struct pass {
	int (*run)(struct pass *, struct ctx *);
	nwalk_fn pre;
	nwalk_fn post;
};


struct symtab syms = { .syms = NULL, .len = 0 };
/* symtab_t locals = { .sym = NULL, .len = 0 }; */

struct ctx *ctx_get(void)
{
	struct ctx *ctx;
	struct prog *prog;

	ctx = calloc(1, sizeof(*ctx));
	ctx->globals = calloc(1, sizeof(*ctx->globals));

	ctx->progs = calloc(3, sizeof(*ctx->progs));

	/* PROBE0 */
	prog = calloc(1, sizeof(*prog));
	prog->locals = calloc(1, sizeof(*prog->locals));
	prog->globals = ctx->globals;

	prog->probe = "k:SyS_read";

	/* { 
	 * 	us = pid();
	 * 	t[0, pid()] = time();
	 * 	reads[pid()] = quantize(arg2);
	 * }
	 */
	prog->ast =
		node_expr(":block",
			  node_expr("=",
				    node_ident("us"),
				    node_expr("pid", NULL),
				    NULL),
			  node_expr("=",
				    node_expr("{}",
					      node_ident("t"),
					      node_num("0"),
					      node_expr("pid", NULL),
					      NULL),
				    node_expr("time", NULL),
				    NULL),
			  node_expr("=",
				    node_expr("{}",
					      node_ident("reads"),
					      node_expr("pid", NULL),
					      NULL),
				    node_expr("quantize", node_ident("arg2"), NULL),
				    NULL),
			  NULL);

	prog->provider = provider_get("k");
	prog->provider->probe(prog);
	/* prog->ir = ir_new(); */
	ctx->progs[0] = prog;

	/* PROBE1 */
	prog = calloc(1, sizeof(*prog));
	prog->locals = calloc(1, sizeof(*prog->locals));
	prog->globals = ctx->globals;

	/* TODO: k -> kret */
	prog->probe = "k:SyS_read2"; 
	/* { times[pid()] = quantize(time() - t[0, pid()]) } */
	prog->ast =
		node_expr("=",
			  node_expr("{}",
				    node_ident("times"),
				    node_expr("pid", NULL),
				    NULL),
			  node_expr("quantize",
				    node_expr("-",
					      node_expr("time", NULL),
					      node_expr("{}",
							node_ident("t"),
							node_num("0"),
							node_expr("time", NULL),
							node_expr("pid", NULL),
							NULL),
					      NULL),
				    NULL),
			  NULL);

	prog->provider = provider_get("k");
	prog->provider->probe(prog);
	/* prog->ir = ir_new(); */
	ctx->progs[1] = prog;

	return ctx;
}


int pass_sym_alloc(struct node *n, void *_prog)
{
	struct prog *prog = _prog;
	struct provider *global = provider_get(":");
	int err = 0;

	switch (n->ntype) {
	case N_EXPR:
	case N_IDENT:
		err = prog->provider->sym_alloc(prog, n);
		if (!err || (err != -ENOENT))
			break;

		err = global->sym_alloc(prog, n);
		break;
	case N_NUM:
	case N_STRING:
		err = global->sym_alloc(prog, n);
	}

	if (err) {
		if ((err == -ENOENT))
			_e("%#N: unknown symbol %N.\n", n, n);
	}

	return err;
}

/* int infer_type_list(struct prog *prog, struct node *n) */
/* { */
/* 	type_t *t; */

/* 	/\* list of lists (code block) => void *\/ */
/* 	if (n->list->ntype == N_LIST) { */
/* 		n->type = &t_void; */
/* 		return 0; */
/* 	} */

/* 	t = n->list->type; */
/* 	if (!t) */
/* 		return 0; */

/* 	switch (t->ttype) { */
/* 	case T_FUNC: */
/* 		n->type = t->t.func.type; */
/* 		break; */
/* 	default: */
/* 		n->type = t; */
/* 	} */
/* 	return 0; */
/* } */

/* int infer_type_keyword(struct prog *prog, struct node *n) */
/* { */
/* 	struct node *dst, *src; */

/* 	switch (n->keyword.class) { */
/* 	case KW_ASSIGN: */
/* 		dst = node_next(n); */
/* 		src = node_next(dst); */
/* 		assert(dst && src); */

/* 		if (!src->type) */
/* 			return 0; */

/* 		/\* TODO: assignment is statement for now. do we need */
/* 		 * c-style assignment expressions? e.g `a = b = 2;` *\/ */
/* 		n->type = &t_void; */

/* 		if (dst->type) */
/* 			return 0; */

/* 		dst->type = src->type; */

/* 		if (dst->ntype != N_IDENT) */
/* 			return 0; */

/* 		return sym_add(dst->sym->st, dst->ident, dst->type, NULL); */

/* 	case KW_BINOP: */
/* 		dst = node_next(n); */
/* 		src = node_next(dst); */
/* 		assert(dst && src); */

/* 		if (!(src->type && dst->type && type_equal(src->type, dst->type))) */
/* 			return 0; */

/* 		n->type = dst->type; */
/* 		return 0; */

/* 	default: */
/* 		n->type = &t_void; */
/* 		return 0; */
/* 	} */

/* 	return -ENOSYS; */
/* } */

/* int infer_type_sym(struct prog *prog, struct node *n) */
/* { */
/* 	struct node *parent, *key; */

/* 	if (n->sym->type) { */
/* 		/\* the symbol type could have been inferred in another */
/* 		 * probe, in that case copy the type to this node. *\/ */
/* 		if (!n->type) */
/* 			n->type = n->sym->type; */

/* 		return 0; */
/* 	} */

/* 	parent = node_up(n); */
/* 	key = node_next(n); */

/* 	/\* match `somemap[somekey]` where the type of the entire */
/* 	 * expression and the type of the key is known, since that */
/* 	 * means the type of the map itself is also known. *\/ */
/* 	if (parent && parent->type */
/* 	    && (parent->list->ntype == N_KEYWORD) */
/* 	    && (parent->list->keyword.class == KW_SUBSCRIPT) */
/* 	    && key && key->type) { */
/* 		n->type = type_map_of(key->type, parent->type); */

/* 		return sym_add(n->sym->st, n->ident, n->type, NULL); */
/* 	} */

/* 	return 0; */
/* } */

int pass_type_infer(struct node *n, void *_prog)
{
	struct prog *prog = _prog;

	if (n->sym->func->type_infer)
		return n->sym->func->type_infer(n->sym->func, n);

	return 0;
}

/* int validate_func(struct node *n) */
/* { */
/* 	struct node *arg; */
/* 	field_t *f; */
/* 	int i; */

/* 	for (arg = node_next(n), f = n->type->t.func.args, i = 1; */
/* 	     arg && f && f->type; arg = node_next(arg), f++, i++) { */
/* 		if (type_compatible(arg->type, f->type)) */
/* 			continue; */

/* 		node_print(n, stderr); */
/* 		fprintf(stderr, ": incompatible type of argument %d (", i); */
/* 		type_dump(arg->type, stderr); */
/* 		fputs("), expected ", stderr); */
/* 		type_dump(f->type, stderr); */
/* 		fputs("\n", stderr); */
/* 		return -EINVAL; */
/* 	} */

/* 	if (!arg && (!f || !f->type)) */
/* 		return 0; */

/* 	if (arg) { */
/* 		node_print(n, stderr); */
/* 		fprintf(stderr, ": too many arguments, expected %d", i); */
/* 		return -EINVAL; */
/* 	} */

/* 	if (f->optional) */
/* 		return 0; */

/* 	node_print(n, stderr); */
/* 	fputs(": too few arguments", stderr); */
/* 	return -EINVAL;	 */
/* } */

/* int pass_validate_types(struct node *n, void *_prog) */
/* { */
/* 	struct prog *prog = _prog; */

/* 	node_print(n, stdout); putchar('\n');  */
/* 	if (!n->type) { */
/* 		node_print(n, stderr); */
/* 		fputs(": type unknown\n", stderr); */
/* 		return -EINVAL; */
/* 	} */

/* 	if (n->ntype != N_LIST) */
/* 		return 0; */

/* 	if (n->list->ntype != N_IDENT) */
/* 		return 0; */

/* 	assert(n->list->type->ttype == T_FUNC); */
/* 	return validate_func(n->list); */
/* } */

/* int validate_syms(struct prog *prog) */
/* { */
/* 	return 0; */
/* } */

/* int run_validate_types(struct pass *pass, struct ctx *ctx) */
/* { */
/* 	struct prog **prog; */
/* 	int err; */

/* 	for (prog = ctx->progs; *prog; prog++) { */
/* 		/\* check syms first to give better error messages. */
/* 		 * e.g. "i: type unknown", not "b-: type unknown" *\/ */
/* 		err = validate_syms(*prog); */
/* 		if (err) */
/* 			return err; */

/* 		err = node_walk((*prog)->ast, pass->pre, pass->post, *prog); */
/* 		if (err) */
/* 			return err; */
/* 	} */

/* 	return 0; */

/* } */

/* int rewrite_const_math(struct node *n) */
/* { */
/* 	int64_t result; */
/* 	struct node *a, *b, *new; */
/* 	int op; */

/* 	/\* TODO: handle L/UL/ULL correctly *\/ */
	
/* 	op = n->list->keyword.op; */
/* 	a = node_next(n->list); */
/* 	b = node_next(a); */

/* 	switch (op) { */
/* 	case '*': result = a->num *  b->num; break; */
/* 	case '/': result = a->num /  b->num; break; */
/* 	case '%': result = a->num %  b->num; break; */
/* 	case '+': result = a->num +  b->num; break; */
/* 	case '-': result = a->num -  b->num; break; */
/* 	case '<': result = a->num << b->num; break; */
/* 	case '>': result = a->num >> b->num; break; */
/* 	default: return 0; */
/* 	} */

/* 	new = node_num(result); */
/* 	new->type = n->type; */
/* 	return node_replace(n, new); */
/* } */

/* int pass_rewrite_ast(struct node *n, void *_prog) */
/* { */
/* 	struct prog *prog = _prog; */
/* 	provider_t *global = provider_get(":"); */
/* 	int err; */

/* 	if (prog->provider->rewrite_node) { */
/* 		err = prog->provider->rewrite_node(prog, n); */
/* 		if (err) */
/* 			return err; */
/* 	} */

/* 	if (global->rewrite_node) { */
/* 		err = global->rewrite_node(prog, n); */
/* 		if (err) */
/* 			return err; */
/* 	} */

/* 	/\* pre-compute binops where both sides are constants *\/ */
/* 	if ((n->ntype == N_LIST) */
/* 	    && (n->list->ntype == N_KEYWORD) */
/* 	    && (n->list->keyword.class == KW_BINOP) */
/* 	    && (node_next(n->list)->ntype == N_NUM) */
/* 	    && (node_next(node_next(n->list))->ntype == N_NUM)) */
/* 		return rewrite_const_math(n); */

/* 	return 0; */
/* } */

/* int generate_ir_ident(struct prog *prog, struct node *n) */
/* { */

/* 	switch (n->sym->type->ttype) { */
/* 	case T_FUNC: */
/* 		return n->sym->type->t.func.generate_ir(prog, n); */
/* 	case T_MAP: */
/* 		ir_emit_ldmap(prog->ir, BPF_REG_0, n->sym); */
/* 		return 0; */
/* 	default: */
/* 		break; */
/* 	} */

/* 	return 0; */
/* } */

/* int pass_generate_ir(struct node *n, void *_prog) */
/* { */
/* 	struct prog *prog = _prog; */

/* 	switch (n->ntype) { */
/* 	case N_LIST: */
/* 		return 0; */
/* 	case N_IDENT: */
/* 		return generate_ir_ident(prog, n); */

/* 	default: */
/* 		break; */
/* 	} */
/* 	return 0; */
/* } */

/* int run_generate_ir(struct pass *pass, struct ctx *ctx) */
/* { */
/* 	struct prog **progp; */
/* 	int err; */

/* 	for (progp = ctx->progs; *progp; progp++) { */
/* 		struct prog *prog = *progp; */

/* 		int return_label = ir_alloc_label(prog->ir); */

/* 		err = prog->provider->ir_prologue ? */
/* 			prog->provider->ir_prologue(prog) : 0; */
/* 		if (err) */
/* 			return err; */

/* 		err = node_walk(prog->ast, NULL, pass_generate_ir, prog); */
/* 		if (err) */
/* 			return err; */

/* 		err = prog->provider->ir_epilogue ? */
/* 			prog->provider->ir_epilogue(prog) : 0; */
/* 		if (err) */
/* 			return err; */

/* 		ir_emit_label(prog->ir, return_label); */
/* 		ir_emit_insn(prog->ir, EXIT, 0, 0); */
/* 	} */

/* 	return 0; */
/* } */

int run_walk(struct pass *pass, struct ctx *ctx)
{
	struct prog **prog;
	int err;

	for (prog = ctx->progs; *prog; prog++) {
		err = node_walk((*prog)->ast, pass->pre, pass->post, *prog);
		if (err)
			return err;
	}

	return 0;
}

struct pass passes[] = {
	{ .run = run_walk, .post = pass_sym_alloc },
	{ .run = run_walk, .post = pass_type_infer },
	/* { .run = run_walk, .post = pass_infer_types }, */
	/* { .run = run_walk, .post = pass_infer_types }, */
	/* { .run = run_validate_types, .post = pass_validate_types }, */
	/* { .run = run_walk, .post = pass_rewrite_ast }, */
	/* { .run = run_generate_ir }, */

	{ NULL }
};

int main(void)
{
	struct ctx *ctx = ctx_get();
	struct prog **prog;
	struct pass *pass;
	int err = 0;

	for (pass = passes; pass->run; pass++) {
		err = pass->run(pass, ctx);
		if (err)
			break;
	}

	for (prog = ctx->progs; *prog; prog++) {
		printf("\n\e[34m%s\e[0m\n", (*prog)->probe);
		ast_fprint(stdout, (*prog)->ast);
		printf("\n-- locals\n");
		symtab_dump((*prog)->locals, stdout);
		/* printf("-- ir\n"); */
		/* ir_dump((*prog)->ir, stdout); */
	}

	printf("\n\n-- globals\n");
	symtab_dump(ctx->globals, stdout);

	/* printf("\n\n-- decls\n"); */
	/* type_dump_decls(stdout); */

	if (err)
		printf("ERR: %d\n", err);

	return err;
}