Re: Reducing unncessary string copying

[Enrico Weigelt <weigelt metux de>]

* André Gillibert <metaentropy gmail com> schrieb:

> This adds some level of indirection to access the data bytes, and some
> space overhead.
> For example, the string "hello" is 6 bytes long (including the zero
> terminator), with GLIBC's malloc, it may occupy 8 bytes (or maybe 16
> bytes).
> With your implementation, it would occupy twice as much memory.

No. With my approach, strings aren't a single bit longer.
References to them require one more byte. (well, maybe more due padding).
On certain platforms/architectures (eg. 64bit), we maybe even could put
that single bit directly into the pointer (using the MSB if we can be
sure the upper half is never used).

> Moreover, the binary code using strings would be larger of a few bytes too.

No. Plain access is not is not a single bit larger, ist just an offset.

My string reference looks like this:

typedef struct
{
    gchar* str;
    unsigned char flags;
}

> And so, once g_strdup() has been called, a new dynamic string is
> created. Consequently, very few strings would be static.

g_strdup() wont be called (directly) very often.
Code that is operating on my GCStry wont use that function, unless
it really wants a physical copy (eg. as buffer, etc)m instead if
uses G_CSTRY_COPY() which only copies non-static strings.

> Storing the string as an array of bytes. The first byte of the array
> would be the "static" bit, and the rest would be the string data.

Yep, also a good idea. Makes the required stack space smaller than
my approach. Perhaps could also be extended to do reference counting.

But: I'm currently lacking sufficient coffeine level to type down
a typedef+macro that a) makes writing string literals easy and
b) make it compile-type typesafe.

Perhaps something like:

typedef {
    unsigned char flags;
    unsigned char str[1];
} GCStr;

#define G_CSTR_LITERAL(s) ("\1" s)


cu
-- 
----------------------------------------------------------------------
 Enrico Weigelt, metux IT service -- http://www.metux.de/

 phone:  +49 36207 519931  email: weigelt metux de
 mobile: +49 151 27565287  icq:   210169427         skype: nekrad666
----------------------------------------------------------------------
 Embedded-Linux / Portierung / Opensource-QM / Verteilte Systeme
----------------------------------------------------------------------