Re: Warnings do include offending filename

[David Gibson <david-xT8FGy+AXnRB3Ne2BGzF6laj5H9X9Tb+@public.gmane.org>]

[-- Attachment #1: Type: text/plain, Size: 4453 bytes --]

On Wed, Feb 01, 2017 at 11:16:54AM +1100, David Gibson wrote:
> On Tue, Jan 31, 2017 at 08:24:48AM +0000, Ian Campbell wrote:
> > On Tue, 2017-01-31 at 10:49 +1100, David Gibson wrote:
> > > On Mon, Jan 30, 2017 at 09:13:05AM +0000, Ian Campbell wrote:
> > > > Hello,
> > > > 
> > > > I wasn't sure how/where to make a wishlist bug report, so I hope this
> > > > will suffice, am happy to be pointed in a different direction though.
> > > > 
> > > > I recently[0] stumbled over around 1,000 of these:
> > > >     Warning (unit_address_vs_reg): Node /soc has a reg or ranges property, but no unit name
> > > >     Warning (unit_address_vs_reg): Node /soc/main-oscillator has a reg or ranges property, but no unit name
> > > >     Warning (unit_address_vs_reg): Node /soc has a reg or ranges property, but no unit name
> > > >     Warning (unit_address_vs_reg): Node /soc has a reg or ranges property, but no unit name
> > > >     Warning (unit_address_vs_reg): Node /soc/main-oscillator has a reg or ranges property, but no unit name
> > > >     Warning (unit_address_vs_reg): Node /soc has a reg or ranges property, but no unit name
> > > >     Warning (unit_address_vs_reg): Node /soc/main-oscillator has a reg or ranges property, but no unit name
> > > > 
> > > > When building the split device tree repo[1] from the Linux source
> > > > (essential it's a build of every single dts in the kernel source).
> > > > 
> > > > The cause of the warning is an issue which needs to be fixed but I
> > > > thought I would mention that it would be very useful (I expect) if dtc
> > > > would include the offending file in warnings (like e.g. gcc would), not
> > > > just because of the number of *.dtb being built here but also due to
> > > > #include and /include/ of .dtsi files.
> > > 
> > > Right, having the filenames - and line numbers - there would certainly
> > > be helpful.  Unfortunately, it's not at all trivial to implement.  As
> > > someone said in a different thread, these checks take place (and have
> > > to) after the tree is completely parsed and we no longer have source
> > > locations readily to hand.
> > 
> > Would it be easier (or possible) to print the name of the eventually-
> > to-be-output binary? At the moment the user is left guessing which one
> > of 1,200 *.dtb files they just built produced each of the similar
> > number of warnings. If the message was instead:
> > 
> >     Warning (unit_address_vs_reg): arch/arm/boot/dts/foo.dtb: Node /soc has a ...
> > 
> > Then that would at least be something to go on.
> > 
> > In fact, given the checks are on the final tree, naming the output file
> >  in the messages seems fairly logical (you could even imagine doing
> > these checks in a separate linter tool after the fact, given the *.dtb
> > as input, I suppose)
> 
> Hm, possible, though a bit messy to do within dtc.  The output file is

s/is/isn't/  duh.

> currently passed into that section of the code, but I guess we could
> add it.
> 
> However, it seems this would more easily be fixed from the Makefile
> side: if you echo a (suitably abbreviated) dtc command line, then it
> should become obvious which dtb the errors are associated with.
> 
> > > It might be possible to attach (optional) srcpos * markers to each
> > > node and property in the live tree.  However, that runs into other
> > > complications: when using dts includes nodes can be merges of
> > > information from several different dtsi files.  So, a node can be
> > > composed of information from several non-contiguous chunks of source
> > > from different files.  It's not clear what should be printed in that
> > > case.
> > 
> > Yeah, tricky!
> > 
> > Hopefully the above suggesting will suffice for 99% of cases.
> 
> Actually, this might be made more feasible by some of the overlay
> stuff in the works.  We're looking at doing the overlay merges in a
> separate pass after parsing is complete.  There are some interesting
> edge cases about whether checks get run before or after that merge,
> but I'm hoping most of the checks can run before.  For those that do,
> we should have a well defined chunk of source from which each node and
> property comes.
> 



-- 
David Gibson			| I'll have my music baroque, and my code
david AT gibson.dropbear.id.au	| minimalist, thank you.  NOT _the_ _other_
				| _way_ _around_!
http://www.ozlabs.org/~dgibson

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 819 bytes --]