Re: [RFC 1/4] ref-filter: start adding strbufs with errors

From: Martin Ågren <martin.agren@gmail.com>
Date: Mar 13, 2018, 19:12 UTC
Message-ID: <CAN0heSoJKxqSb0q5SR2SVfEzh6QjeiNNnESorZyYHTSC71jZuQ@mail.gmail.com>
In-Reply-To: <010201621edc97a4-a4239b1e-86d2-4400-aaac-d81727710f4f-000000@eu-west-1.amazonses.com>

On 13 March 2018 at 11:16, Olga Telezhnaya <olyatelezhnaya@gmail.com> wrote:

Show 10 quoted lines

> This is a first step in removing any printing from
> ref-filter formatting logic, so that it could be more general.
> Everything would be the same for show_ref_array_item() users.
> But, if you want to deal with errors by your own, you could invoke
> format_ref_array_item(). It means that you need to print everything
> (the result and errors) on your side.
>
> This commit changes signature of format_ref_array_item() by adding
> return value and strbuf parameter for errors, and fixes
> its callers.

Minor nit: Maybe s/fixes its callers/adjusts its callers/. They are not broken or need to be fixed. They were simply playing the game according to the old rules, and now they need to learn the new ways. :-)

Show 16 quoted lines

> Signed-off-by: Olga Telezhnaia <olyatelezhnaya@gmail.com>
> ---
>  builtin/branch.c |  7 +++++--
>  ref-filter.c     | 17 ++++++++++++-----
>  ref-filter.h     |  7 ++++---
>  3 files changed, 21 insertions(+), 10 deletions(-)
>
> diff --git a/builtin/branch.c b/builtin/branch.c
> index 8dcc2ed058be6..f86709ca42d5e 100644
> --- a/builtin/branch.c
> +++ b/builtin/branch.c
> @@ -391,7 +391,6 @@ static void print_ref_list(struct ref_filter *filter, struct ref_sorting *sortin
>         struct ref_array array;
>         int maxwidth = 0;
>         const char *remote_prefix = "";
> -       struct strbuf out = STRBUF_INIT;

You move this variable into the loop to reduce its scope. At first I suspected that this might mean we now start allocating+releasing in each run of the loop, which might be a performance-regression. But it turns out, we already did that, so this tightening of the scope has no such downsides. :-) From the commit message, I wasn't expecting this move, though. Maybe "While at it, reduce the scope of the out-variable."

Show 12 quoted lines

>         char *to_free = NULL;
>
>         /*
> @@ -419,7 +418,10 @@ static void print_ref_list(struct ref_filter *filter, struct ref_sorting *sortin
>         ref_array_sort(sorting, &array);
>
>         for (i = 0; i < array.nr; i++) {
> -               format_ref_array_item(array.items[i], format, &out);
> +               struct strbuf out = STRBUF_INIT;
> +               struct strbuf err = STRBUF_INIT;
> +               if (format_ref_array_item(array.items[i], format, &out, &err))
> +                       die("%s", err.buf);

Using "%s", good.

Show 54 quoted lines

>                 if (column_active(colopts)) {
>                         assert(!filter->verbose && "--column and --verbose are incompatible");
>                          /* format to a string_list to let print_columns() do its job */
> @@ -428,6 +430,7 @@ static void print_ref_list(struct ref_filter *filter, struct ref_sorting *sortin
>                         fwrite(out.buf, 1, out.len, stdout);
>                         putchar('\n');
>                 }
> +               strbuf_release(&err);
>                 strbuf_release(&out);
>         }
>
> diff --git a/ref-filter.c b/ref-filter.c
> index 45fc56216aaa8..54fae00bdd410 100644
> --- a/ref-filter.c
> +++ b/ref-filter.c
> @@ -2118,9 +2118,10 @@ static void append_literal(const char *cp, const char *ep, struct ref_formatting
>         }
>  }
>
> -void format_ref_array_item(struct ref_array_item *info,
> +int format_ref_array_item(struct ref_array_item *info,
>                            const struct ref_format *format,
> -                          struct strbuf *final_buf)
> +                          struct strbuf *final_buf,
> +                          struct strbuf *error_buf)
>  {
>         const char *cp, *sp, *ep;
>         struct ref_formatting_state state = REF_FORMATTING_STATE_INIT;
> @@ -2148,19 +2149,25 @@ void format_ref_array_item(struct ref_array_item *info,
>                 resetv.s = GIT_COLOR_RESET;
>                 append_atom(&resetv, &state);
>         }
> -       if (state.stack->prev)
> -               die(_("format: %%(end) atom missing"));
> +       if (state.stack->prev) {
> +               strbuf_addstr(error_buf, _("format: %(end) atom missing"));
> +               return -1;
> +       }
>         strbuf_addbuf(final_buf, &state.stack->output);
>         pop_stack_element(&state.stack);
> +       return 0;
>  }
>
>  void show_ref_array_item(struct ref_array_item *info,
>                          const struct ref_format *format)
>  {
>         struct strbuf final_buf = STRBUF_INIT;
> +       struct strbuf error_buf = STRBUF_INIT;
>
> -       format_ref_array_item(info, format, &final_buf);
> +       if (format_ref_array_item(info, format, &final_buf, &error_buf))
> +               die("%s", error_buf.buf);
>         fwrite(final_buf.buf, 1, final_buf.len, stdout);
> +       strbuf_release(&error_buf);

I think this `strbuf_release()` will never actually do anything. If we get here, we had no error. But it makes sense (to me) to always be clear about releasing this. In this case it is easy enough.

Possible counterargument: We might want this sort of "error-handling by strbufs" to follow this simple rule: return an error if and only if you add some error-string to the buffer. If this rule is universal enough, it might be ok to skip releasing these sort of buffers if you do not have an error.

Martin