[
https://issues.apache.org/jira/browse/MESOS-3214?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Michael Park updated MESOS-3214:
--------------------------------
Description:
It's desirable to replace the boost {{foreach}} macro with the C++11
range-based {{for}}. This will help avoid some of the pitfalls of boost
{{foreach}} such as dealing with types with commas in them, as well as
improving compiler diagnostics by avoiding the macro expansion.
One way to accomplish this is to replace the existing {{foreach (const Elem&
elem, container)}} pattern with {{for (const Elem& elem : container)}}. We
could support {{foreachkey}} and {{foreachvalue}} semantics via adaptors
{{keys}} and {{values}} which would be used like this: {{for (const Key& key :
keys(container))}}, {{for (const Value& value : values(container))}}. This
leaves {{foreachpair}} which cannot be used with {{for}}. I think it would be
desirable to support {{foreachpair}} for cases where the implicit unpacking is
useful.
Another approach is to keep {{foreach}}, {{foreachpair}}, {{foreachkey}} and
{{foreachvalue}}, but simply implement them based on range-based {{for}}. For
example, {{#define foreach(elem, container) for (elem : container)}}. While the
consistency in the names is desirable, but unnecessary indirection of the macro
definition is not.
It's unclear to me which approach we would favor in Mesos, so please share your
thoughts and preferences.
was:
Replace boost {{foreach}} macro with the C++11 range-based {{for}}. This will
help avoid some of the pitfalls of boost {{foreach}} such as dealing with types
with commas in them, as well as improving compiler diagnostics by avoiding the
macro expansion.
The existing {{foreach (const Elem& elem, container)}} pattern can be replaced
with {{for (const Elem& elem : container)}}.
{{foreachpair}}, {{foreachkey}} and {{foreachvalue}} will still be supported
for cases where the implicit unpacking is useful.
The implementation of {{foreachpair}} can be simplified with the use of
range-based for within, {{foreachkey}} and {{foreachvalue}} will be exactly as
is except it can use {{std::ignore}} instead of the hand-rolled version.
> Replace boost foreach with range-based for
> ------------------------------------------
>
> Key: MESOS-3214
> URL: https://issues.apache.org/jira/browse/MESOS-3214
> Project: Mesos
> Issue Type: Task
> Components: stout
> Reporter: Michael Park
> Labels: mesosphere
>
> It's desirable to replace the boost {{foreach}} macro with the C++11
> range-based {{for}}. This will help avoid some of the pitfalls of boost
> {{foreach}} such as dealing with types with commas in them, as well as
> improving compiler diagnostics by avoiding the macro expansion.
> One way to accomplish this is to replace the existing {{foreach (const Elem&
> elem, container)}} pattern with {{for (const Elem& elem : container)}}. We
> could support {{foreachkey}} and {{foreachvalue}} semantics via adaptors
> {{keys}} and {{values}} which would be used like this: {{for (const Key& key
> : keys(container))}}, {{for (const Value& value : values(container))}}. This
> leaves {{foreachpair}} which cannot be used with {{for}}. I think it would be
> desirable to support {{foreachpair}} for cases where the implicit unpacking
> is useful.
> Another approach is to keep {{foreach}}, {{foreachpair}}, {{foreachkey}} and
> {{foreachvalue}}, but simply implement them based on range-based {{for}}. For
> example, {{#define foreach(elem, container) for (elem : container)}}. While
> the consistency in the names is desirable, but unnecessary indirection of the
> macro definition is not.
> It's unclear to me which approach we would favor in Mesos, so please share
> your thoughts and preferences.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
