ast, eval, lexer: keep track of overflows when parsing numbers

This allows number literals that exceed the range INT64_MIN..INT64_MAX
to be truncated to the respective min and max values in a defined manner.

It also makes it possible to have the expression `{{ -9223372036854775808 }}`
actually result in `-9223372036854775808`. Since negation and number
declaration are separate operations, the value would be first truncated to
`9223372036854775807` and then negated, making it impossible to write a
literal INT64_MIN value without tracking the overflow.

Also fix the number parsing logic to not trucate intergers to 32bit.

Signed-off-by: Jo-Philipp Wich <jo@mein.io>

1 files changed, 1 insertions, 0 deletions

diff --git a/ast.h b/ast.h
index b3dfa3b..b45ee41 100644
--- a/ast.h
+++ b/ast.h
@@ -54,6 +54,7 @@ struct ut_op {
 	uint16_t type;
 	uint16_t is_first:1;
 	uint16_t is_op:1;
+	uint16_t is_overflow:1;
 	uint32_t off;
 	struct json_object *val;
 	union {