Re: [PATCH 05/16] replace-object: eliminate replace objects prepared flag

[René Scharfe]

Am 10.04.2018 um 00:45 schrieb Stefan Beller:
> By making the oidmap a pointer, we eliminate the need for
> the global boolean variable 'replace_object_prepared'.
> 
> Signed-off-by: Stefan Beller 
> ---
>   object-store.h   |  2 +-
>   replace-object.c | 16 +---
>   2 files changed, 10 insertions(+), 8 deletions(-)
> 
> diff --git a/object-store.h b/object-store.h
> index c04b4c95eb..1ff862c7f9 100644
> --- a/object-store.h
> +++ b/object-store.h
> @@ -99,7 +99,7 @@ struct raw_object_store {
>* Objects that should be substituted by other objects
>* (see git-replace(1)).
>*/
> - struct oidmap replace_map;
> + struct oidmap *replace_map;

This also allows the '#include "oidmap.h"' introduced in patch 3 to be
replaced by 'struct oidmap;' (forward declaration instead of include).
Keeping the type opaque discourages circumventing accessor functions;
not dragging in other headers avoids some compile time overhead.

René

Re: [PATCH 05/16] replace-object: eliminate replace objects prepared flag

[Junio C Hamano]

Stefan Beller  writes:

> By making the oidmap a pointer, we eliminate the need for
> the global boolean variable 'replace_object_prepared'.

That is not quite a justification for this change, as making it a
pointer (and paying for the malloc(3) overhead) is not the only way
to remove the variable (i.e. the "has this been initialized?" bit
can be moved to "struct raw_object_store").

One possible advantage of this approach, I guess, is that we would
more quickly catch code that tries to access replace-map without
initializing it.