Re^6: Bidirectional lookup algorithm? (Judy)

[oiskuu]

A long long typedef would break your assumption on x32 ABI. Fortunately, there is a type that should fit the bill exactly. You can use typedef ptrdiff_t Word_t; (and ssize_t is almost always the same size as well).

$ echo | gcc -mx32 -dM -E - | grep SIZEOF

Funny about the JU_LEASTBYTESMASK macro—it still doesn't appear quite right. Diligently the BITSPERBYTE is incorporated, but then, in the same definition, the 0x100 constant assumes 8 bits per byte ...

[BrowserUk]

A long long typedef would break your assumption on x32 ABI.

I meant (and thought I'd implied) "on 64-bit systems".

But you are right, there are probably better choices.

Fortunately, there is a type that should fit the bill exactly. You can use typedef ptrdiff_t Word_t;

ptrdiff_t is described as: "Result of subtraction of two pointers.".

Which means it is a signed type, which might have implications for the kind of bit manipulations -- shifts and masking -- the judy code performs.

uintptr_t would probably be a better choice: "uintptr_t: An unsigned integer or unsigned __int64 version of intptr_t.".

(and ssize_t is almost always the same size as well).

There doesn't appear to be a ssize_t defined by the MS compiler.

There's size_t defined as:"Result of sizeof operator.", (as used by Perl's STRLEN type), which seems to happily coexist with any of U32/I32 or U64/I64 variables as counters in your typical for loop.

[syphilis]

Funny about the JU_LEASTBYTESMASK macro—it still doesn't appear quite right

Yes, and it seems to have some involvement in the errors that kill the build.
In JudyInsArray.c we have:

static Word_t subexp_mask[] = {
    0,
    ~cJU_POP0MASK(1),
    ~cJU_POP0MASK(2),
    ~cJU_POP0MASK(3),
#ifdef JU_64BIT
    ~cJU_POP0MASK(4),
    ~cJU_POP0MASK(5),
    ~cJU_POP0MASK(6),
    ~cJU_POP0MASK(7),
#endif

There's no problem with the first 5 elements listed there, but in relation to ~cJU_POP0MASK(5), ~cJU_POP0MASK(6) and ~cJU_POP0MASK(7) we get the error that they are not constant expressions ("initializer element is not constant"), thus breaking C's rules.

And in JudyCommon/JudyPrivateBranch.h we find:

#define cJU_POP0MASK(cPopBytes) JU_LEASTBYTESMASK(cPopBytes)

Cheers,
Rob