[lldb-dev] Changes to Host incoming

[Zachary Turner]

So as most of you know, I've been working to get Windows stuff up and
running.  An important part of this involves creating a robust Host layer
for Windows.  Anybody who's looked at Host.cpp knows it's a mess of #ifdefs
and platform-specific logic, and with Windows this is going to get much
worse as I fill out more and more of Host.cpp.  Many things on Windows have
no analogue on other platforms, some things have an analogue but not a very
good one, some things have a partial analogue, etc.  After a few attempts
to implement some things in Host.cpp I decided it wasn't going to work, or
at the very least it wasn't going to be maintainable.  And I don't want my
work on Windows to affect other peoples' ability to get the things they
need to get done, done.

I submitted a patch last week to move all host FileSystem related code to a
separate FileSystem class, and in a few minutes I'm going to post a patch
that starts a more significant refactor of Host.  The idea here is to
(maybe) get rid of the Host class entirely, and instead move functionality
in classes that are more specific and descriptive of their purpose.
 FileSystem was the start of this, and the upcoming patch will implement
HostInfo - a static class used only for answering queries about the host
operating system.  Future patches will split out code for manipulating
processes and threads.

There are, at least in my view, a number of benefits to doing things this
way:

1) Maintaining code in the Host layer becomes easier.  Very few (if any)
#ifdefs will remain at all.
2) Better logical separation of responsibilities.  There was not much rhyme
or reason before as to what kind of functionality would be behind Host::,
and what would be in some other class such as Pipe, or Socket.
3) Less risk of breaking other platforms when you make platform-specific
changes for your own Host.  Adding a new function for MacOSX?  You'll most
likely be editing source/Host/macosx/HostInfoMacOSX.mm.  Zero chance of
breaking another platform.

There is one aspect which may or may not be contentious, and that is that
I've opted to omit default implementations of functions.  As a concrete
example, previously we had the function Host::GetDistribution.  This is now
implemented in HostInfoLinux::GetDistribution().  It is not implemented
anywhere else.  If you write HostInfo::GetDistribution() while not on
Linux, you will get a compile error.  This forces you to put the platform
specific checks at the call-site.  I actually think this is a benefit,
because if a function is truly platform specific then you should be calling
it from a codepath that is aware of the platform.

I'm open to hearing other viewpoints about this though.

The patch will be up shortly.  It's large, so I hope this email serves to
give a summary of its intent.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/lldb-dev/attachments/20140818/4245a7af/attachment.html>