7 ms·

The Most Expensive One-Byte Mistake: Was Nul-Terminated Text Strings Wrong?

12 points by pcr910303 7y ago

olliej 7y ago
Can we add (2011) to the title?
downerending 7y ago
This would be a lot more compelling if the author also listed the benefits of nul-terminated strings.
- olliej 7y ago
  The only benefit is a very slight reduction in memory usage. Even that is up for debate as the cost of computing the length of a null terminated string is so high that variable length integer for the size is going to typically be the same space, and even when not will still be faster. Null termination was easy to implement quickly. It’s bad everywhere else.
  - downerending 7y ago
    Guess this would make a good interview question. Another benefit is that you don't have to decide on a maximum string length. Initially, they probably would have chosen a uint16 for the length, but over time we would have had to add uint32 and uint64. Old code doesn't completely fade away, so we would have had three similar but not-quite-compatible string types for all time. A small further benefit is that strings with the same suffix can share the same storage. I suspect there's more that I'm missing.
    - olliej 7y ago
      Variable length integers are used in many places (protobuf, utf, etc), so you could also have unbound length there as well. My 100% perfect concept ( :D ) would be [length]<here's the pointer>[data]where length would be read high to low. Because of the functionally upper capped length (maximum address space) we can treat accessing that as O(1) vs. O(N) for null termination, etc.
      - 7786655 7y ago
        You're overcomplicating the problem. void accept_string(const char *a, size_t a_length); is good enough. It has the additional advantage that you can give it a substring out of a larger string without copying it.
perl4ever 7y ago
Null terminated strings always seemed to me like a terrible idea, but on the other hand, 255-character maximums aren't really reasonable. What we need are strings prefixed with a length of the length of the string and then the length as a variable length integer.

12 points by pcr910303 7y ago

olliej 7y ago

Can we add (2011) to the title?

downerending 7y ago

This would be a lot more compelling if the author also listed the benefits of nul-terminated strings.

olliej 7y ago
The only benefit is a very slight reduction in memory usage. Even that is up for debate as the cost of computing the length of a null terminated string is so high that variable length integer for the size is going to typically be the same space, and even when not will still be faster. Null termination was easy to implement quickly. It’s bad everywhere else.
- downerending 7y ago
  Guess this would make a good interview question. Another benefit is that you don't have to decide on a maximum string length. Initially, they probably would have chosen a uint16 for the length, but over time we would have had to add uint32 and uint64. Old code doesn't completely fade away, so we would have had three similar but not-quite-compatible string types for all time. A small further benefit is that strings with the same suffix can share the same storage. I suspect there's more that I'm missing.
  - olliej 7y ago
    Variable length integers are used in many places (protobuf, utf, etc), so you could also have unbound length there as well. My 100% perfect concept ( :D ) would be [length]<here's the pointer>[data]where length would be read high to low. Because of the functionally upper capped length (maximum address space) we can treat accessing that as O(1) vs. O(N) for null termination, etc.
    - 7786655 7y ago
      You're overcomplicating the problem. void accept_string(const char *a, size_t a_length); is good enough. It has the additional advantage that you can give it a substring out of a larger string without copying it.

perl4ever 7y ago

Null terminated strings always seemed to me like a terrible idea, but on the other hand, 255-character maximums aren't really reasonable. What we need are strings prefixed with a length of the length of the string and then the length as a variable length integer.