`strcmp` falsely claims it returns a 0, 1 or -1

[ash-m]

The docs for strcmp say:

Return Values ¶

Returns -1 if string1 is less than string2; 1 if string1 is greater than string2, and 0 if they are equal.

Changelog ¶

Version	Description
8.2.0	This function now returns -1 or 1, where it previously returned a negative or positive integer.

Problems:

• The "Return Values" part is patentenly not true: https://3v4l.org/CXhkE#v8.2.0
• I believe the "Changelog" is refering to when a string is compared to an empty string: https://3v4l.org/UHXcr

I believe the intended logic here is:

• Attempt to calculate the spread between to strings (unpredictable for large spreads)
• If the strings match except for trailing characters, return the difference in characters

Whatever the logic, it seems like it does not return very predicable results other than being contrasted to 0. Here are some interesting return values: https://3v4l.org/FN86s

[cmb69]

See #3223. While this has been fixed in the migration guide (but not in UPGRADING), the fix apparently missed the changelog entries.

[mickmackusa]

In addition to strcmp() and substr_compare() previously mentioned, the following can be added to the set of misleading documentations:

• strcasecmp()
• strncmp()
• strncasecmp()

https://3v4l.org/Kumk4

[IMSoP]

I don't think there's any intent for these return values to be predictable or meaningful, only useful as a callback for functions like usort. The documentation should simply say:

Returns a value less than 0 if string1 is less than string2; a value more than 0 if string1 is greater than string2, and 0 if they are equal.

[kamil-tekiela]

I don't think there's any intent for these return values to be predictable or meaningful, only useful as a callback for functions like usort. The documentation should simply say:

Returns a value less than 0 if string1 is less than string2; a value more than 0 if string1 is greater than string2, and 0 if they are equal.

To be honest, I didn't even know they had any meaning. I thought it was just a random number.

[IMSoP]

To be honest, I didn't even know they had any meaning. I thought it was just a random number.

I think the accurate word would be arbitrary rather than random: there is no attempt to avoid meaning, the values are just a side effect of the underlying implementation. Specifically, they are probably the result of subtracting one value from another, which is generally an efficient way of comparing them - if the result is 0, they're equal; otherwise, the sign of the result indicates which is larger.

[cmb69]

I agree with @IMSoP. But see php/php-src#8220 which introduced the incomplete normalization to -1/0/1.

[drealecs]

Maybe we should just revert #1857, as this might be the wrong one.
The changes were intended just for length comparison: https://3v4l.org/dLXlq

[mickmackusa]

When used as the callback of array_uintersect*() or array_udiff*(), is there a performance benefit to the under-the-hood sorting if the return values of strcmp() (and other related comparison functions) can be beyond -1, 0, 1? Are there potentially fewer ties to break if a greater range of positive and negative return values is possible? Mick