On Wed, Jan 4, 2017 at 6:22 PM, William A Rowe Jr <[email protected]> wrote:
> On Wed, Jan 4, 2017 at 11:12 AM, Yann Ylavic <[email protected]> wrote:
>>
>> This would work for me (on the proxy side), too.
>> The patch (attached) is a bit longer, but still reasonable IMHO.
>> WDYT?
>
> Not understanding if (!header->key) { continue; } - why success if there is
> a dead ': UnnamedValue' entry in the output headers table?
That's quite due to apr_table internals, it seems that
apr_table_entry_t->key can be set to NULL there to remove an entry.
Every apr_table_elts() iterator I've seen or worked with does this
check, and so does apr_table_[v]do BTW.
> Unclear why
> apr_table_do needed to be re-implemented.
Because we can't change the value in an apr_table_do callback.
(Note this patch won't change the value in place, but will make it
point to another buffer).
>
> Does it make sense to smash surrounding whitespace here? A single SP
> is going to have the same semantic value as multiple SP/TAB characters.
Makes sense, will do it on commit ("end += 3 + strspn(end + 3, "\t
");") if no one is against the whole change.
