Web lists-archives.com

Re: [PATCH v3 5/8] evolve: Add the change-table structure




Works with me. I'll resubmit without the double casting next chance I
have to work on it. My long-term plan for that struct was to use the
memory pool for all allocations anyway. I think that should let me
implement it without moving objects around, which will make their
addresses stable. That should let me use pointers for everything,
without the ints - so I probably won't need the union.

On Tue, Jan 29, 2019 at 10:02 AM Junio C Hamano <gitster@xxxxxxxxx> wrote:
>
> Stefan Xenos <sxenos@xxxxxxxxxx> writes:
>
> > Sorry, folks. I normally can't do any open source work on weekdays.
> > That also includes writing responses on the mailing list, so there
> > will normally be a week or two lag for me to iterate on this sort of
> > thing.
> >
> > Feel free to either include this fix or revert my patch if there's a
> > problem with it - just let me know what you selected. I'll roll with
> > it and either resubmit with the requested changes or submit the
> > requested changes as follow-ups.
>
> I think double casting Dscho did was not his ideal "fix" but he did
> it as a mere workaround to force the 'pu' branch to compile.  And I
> also agree with him that the double casting workaround is too ugly
> to live, compared to a union he suggests.  So I'd rather kick the
> branch out of 'pu' for now.
>
> Thanks, both.
>
> >
> >   - Stefan
> >
> > On Mon, Jan 28, 2019 at 3:08 PM Johannes Schindelin
> > <Johannes.Schindelin@xxxxxx> wrote:
> >>
> >> Hi Junio,
> >>
> >> On Mon, 28 Jan 2019, Johannes Schindelin wrote:
> >>
> >> > On Sun, 27 Jan 2019, sxenos@xxxxxxxxxx wrote:
> >> >
> >> > > +   new_item->util = (void*)index;
> >> >
> >> > This is not good. You are using a `long` here. The 80s called and want
> >> > their now-obsolete data types back.
> >> >
> >> > If you want a data type that can take an integer but also a pointer, use
> >> > `intptr_t` instead.
> >> >
> >> > But even that is not good practice. What you really want here is to use a
> >> > union of the data types that you want to store in that `util` field.
> >> >
> >> > This is not merely academic, your code causes compile errors on Windows:
> >> >
> >> > https://dev.azure.com/gitgitgadget/git/_build/results?buildId=400&view=logs&jobId=fd490c07-0b22-5182-fac9-6d67fe1e939b&taskId=ce91d5d6-0c55-50f5-8ab9-6695c03ab102&lineStart=430&lineEnd=440&colStart=1&colEnd=1
> >>
> >> Since Stefan did not grace us with an answer, Junio, could I ask you to
> >> squash this in (which is by no means a satisfactory fix, but it is a
> >> stopgap to get `pu` building again)?
> >>
> >> -- snipsnap --
> >> diff --git a/change-table.c b/change-table.c
> >> index 2e0d935de846..197ce2783532 100644
> >> --- a/change-table.c
> >> +++ b/change-table.c
> >> @@ -103,7 +103,7 @@ void change_table_add(struct change_table *to_modify, const char *refname,
> >>         new_head->hidden = starts_with(refname, "refs/hiddenmetas/");
> >>
> >>         new_item = string_list_insert(&to_modify->refname_to_change_head, refname);
> >> -       new_item->util = (void*)index;
> >> +       new_item->util = (void *)(intptr_t)index;
> >>         // Use pointers to the copy of the string we're retaining locally
> >>         refname = new_item->string;
> >>
> >> @@ -201,6 +201,6 @@ struct change_head* get_change_head(struct change_table *heads,
> >>                 return NULL;
> >>         }
> >>
> >> -       index = (long)item->util;
> >> +       index = (long)(intptr_t)item->util;
> >>         return &(heads->heads.array[index]);
> >>  }
> >>