Hi Philippe
You wrote:
Declaring the last member as char[1] is not the best choice. char[0] would be better.
It seems to me [0] isn't allowed by C90 and C99 specs.
Extracts from this article:
- declaring zero-length arrays is allowed in GNU C as an extension
- in ISO C90 the
- ISO C99 introduces flexible array member : the array is declared without a size, just []
Do you advocate zero length only to optimize the size of the struct ?
What about its compatibility with the most popular compilers ?
M
The safe option that does it waste extra padding bytes at end of the structure is to declare the last byte array with a size 0 and not one. The compiler will align the byte array if needed it it should not do that, because byte arrays should be byte aligned and not word aligned except if this array has some prepaddng declared or if the array has specific declarators to force the alignment (including by de Mari g this array in an union with other types with alignment constraints)
Declaring the last member as char[1] is not the best choice. char[0] would be better.
> typedef struct TString {
> CommonHeader;
> lu_byte extra;
> lu_byte shrlen;
> unsigned int hash;
> union {
> size_t lnglen;
> struct TString *hnext;
> } u;
> char bytes[1];
> } TString;
>
> #define getstr(ts) check_exp(sizeof((ts)->extra), ts->bytes)
>
> #define sizelstring(l) (sizeof(TString) + (l) * sizeof(char))
This code wastes up to seven bytes (due to alignment), compared with
the original version. A better option would be to use 'offsetof' to
compute the size of the rest of the structure:
#define sizelstring(l) (offsetof(TString, bytes) + ((l) + 1) * sizeof(char))
I hope the new gcc does not complain about that code.
-- Roberto
_______________________________________________
lua-l mailing list -- lua-l@lists.lua.org
To unsubscribe send an email to lua-l-leave@lists.lua.org
lua-l mailing list -- lua-l@lists.lua.org
To unsubscribe send an email to lua-l-leave@lists.lua.org
