Re: [Wireshark-commits] master 9079e3a: Cheat and try to fix the generated file manually.

[Evan Huus <eapache () gmail com>]

On Mon, Jun 23, 2014 at 4:32 PM, Pascal Quantin <pascal.quantin () gmail com>
wrote:

> Hi all,
>
> Le 23/06/2014 22:22, Jakub Zawadzki a écrit :
> > Hello Evan,
> >
> > On Mon, Jun 23, 2014 at 02:10:13PM -0400, Evan Huus wrote:
> > > Storing generated files in source control makes maintenance and patch
> > > review much harder and puts extra requirements on us to keep things in
> > > sync. I'd rather the computer do the extra work :)
> > But in the same time, we can have 3rd party library, like nghttp2[1]
> (12KLOC) put into git.
> > This is already more than 30 commits behind nghttp2/lib git HEAD,
> > nor README.nghttp2 is updated about recent fixes in this forked code.
> FYI the only fixes done on top of those listed in README.nghttp2 should
> be casts to ensure proper compilation on Windows. AFAIK Alexis intends
> to submit those changes upstream. And we are only interested by the
> decompression part of the library.
> We decided to add nghttp2 to Wireshark source code (for now) as it is
> not available to all platforms yet. We might remove it in the future and
> use 3rd party packages if required. I still consider this as a WIP and
> nothing is frozen. But at least it provides decompression functionality
> for the still moving HTTP2 drafts.
>
> > > We store C files and require a compiler, where we could just store the
> > > compiled .o files in source control. To me this is much the same thing.
> > Great to know.
> IMHO, storing source files (even if those are the output of generation
> tools) can hardly be compared as storing object files.

I went looking on google for opinions and arguments on this topic. A
surprising number of sites actually suggested just this: storing the
complete compiled executable and all objects in source control. I no longer
claim to understand software development best practices :/


> > > Balint and Jörg and I discussed that we would eventually like to remove
> the
> > > ASN.1 generated source from git as well, and make that a standard part
> of
> > > the build process.
> > I am happy to hear this, please do.
> > For sure you want also do this for X11 dissector, which for rebuilt only
> requires
> > downloading mesa sources.
> While I understand the objective, we must keep in mind that removing the
> ASN.1 generated files will increase by a non negligible time the first
> build time (which means all buildbot builds). It represents severals
> minutes on my machine....

Something to be aware of, but I don't think the buildbots are under
sufficient load for this to be a serious problem. Once the build system is
fixed to permit parallel ASN.1 compilation that time should go down quite a
bit as well.

___________________________________________________________________________
Sent via:    Wireshark-dev mailing list <wireshark-dev () wireshark org>
Archives:    http://www.wireshark.org/lists/wireshark-dev
Unsubscribe: https://wireshark.org/mailman/options/wireshark-dev
             mailto:wireshark-dev-request () wireshark org?subject=unsubscribe