A question of sstate and compilers

A question of sstate and compilers

[Gary Thomas]

I'm trying to isolate a problem that showed up in my builds
in the last month.  Little has changed in the sources, but
the one big change was I moved from GCC/4.9 to GCC/5.x  On
the surface, this seemed to be a non-consequence, but I have
one very subtle corner case that is now broken.  In an effort
to isolate the issue (I no longer think it was the compiler
change), I went back to GCC/4.9.  This caused my build (tree)
which has existed for many months (the same build tree started
in Feb 2016) to basically rebuild everything.

There's the rub - shouldn't the sstate-cache hold all of those
old bits and just be able to re-stage?  I'm a bit confused about
that.  Even worse, I switched back to GCC/5.x, didn't touch
anything else in my sources or build tree, and now it's off
again, [re]building the majority of my packages.

Any explanations?  Is this correct behavior?

-- 
------------------------------------------------------------
Gary Thomas                 |  Consulting for the
MLB Associates              |    Embedded world
------------------------------------------------------------

Re: A question of sstate and compilers

[Burton, Ross]

[-- Attachment #1: Type: text/plain, Size: 1117 bytes --]

On 30 July 2016 at 09:41, Gary Thomas <gary@mlbassoc.com> wrote:

> I'm trying to isolate a problem that showed up in my builds
> in the last month.  Little has changed in the sources, but
> the one big change was I moved from GCC/4.9 to GCC/5.x  On
> the surface, this seemed to be a non-consequence, but I have
> one very subtle corner case that is now broken.  In an effort
> to isolate the issue (I no longer think it was the compiler
> change), I went back to GCC/4.9.  This caused my build (tree)
> which has existed for many months (the same build tree started
> in Feb 2016) to basically rebuild everything.
>
> There's the rub - shouldn't the sstate-cache hold all of those
> old bits and just be able to re-stage?  I'm a bit confused about
> that.  Even worse, I switched back to GCC/5.x, didn't touch
> anything else in my sources or build tree, and now it's off
> again, [re]building the majority of my packages.
>

It's possible that there was some other change that caused a rebuild.
 bitbake-whatchanged, bitbake -S printdiff, etc will help you chase this if
you want to.

Ross

[-- Attachment #2: Type: text/html, Size: 1578 bytes --]

Re: A question of sstate and compilers

[Gary Thomas]

On 2016-08-01 16:04, Burton, Ross wrote:
>
> On 30 July 2016 at 09:41, Gary Thomas <gary@mlbassoc.com <mailto:gary@mlbassoc.com>> wrote:
>
>     I'm trying to isolate a problem that showed up in my builds
>     in the last month.  Little has changed in the sources, but
>     the one big change was I moved from GCC/4.9 to GCC/5.x  On
>     the surface, this seemed to be a non-consequence, but I have
>     one very subtle corner case that is now broken.  In an effort
>     to isolate the issue (I no longer think it was the compiler
>     change), I went back to GCC/4.9.  This caused my build (tree)
>     which has existed for many months (the same build tree started
>     in Feb 2016) to basically rebuild everything.
>
>     There's the rub - shouldn't the sstate-cache hold all of those
>     old bits and just be able to re-stage?  I'm a bit confused about
>     that.  Even worse, I switched back to GCC/5.x, didn't touch
>     anything else in my sources or build tree, and now it's off
>     again, [re]building the majority of my packages.
>
>
> It's possible that there was some other change that caused a rebuild.  bitbake-whatchanged, bitbake -S printdiff, etc
> will help you chase this if you want to.

Nothing else changed, I'm sure.

-- 
------------------------------------------------------------
Gary Thomas                 |  Consulting for the
MLB Associates              |    Embedded world
------------------------------------------------------------

Re: A question of sstate and compilers

[Burton, Ross]

[-- Attachment #1: Type: text/plain, Size: 261 bytes --]

On 1 August 2016 at 15:26, Gary Thomas <gary@mlbassoc.com> wrote:

> Nothing else changed, I'm sure.
>

*Something* changed, although it may be something that shouldn't have
caused a rebuild (at which point printdiff is useful to find that out).

Ross

[-- Attachment #2: Type: text/html, Size: 659 bytes --]

Re: A question of sstate and compilers

[Gary Thomas]

On 2016-08-01 16:28, Burton, Ross wrote:
>
> On 1 August 2016 at 15:26, Gary Thomas <gary@mlbassoc.com <mailto:gary@mlbassoc.com>> wrote:
>
>     Nothing else changed, I'm sure.
>
>
> *Something* changed, although it may be something that shouldn't have caused a rebuild (at which point printdiff is
> useful to find that out).

I'll try running this and report, but the only thing that changed
is the line in local.conf selecting the version of GCC

-- 
------------------------------------------------------------
Gary Thomas                 |  Consulting for the
MLB Associates              |    Embedded world
------------------------------------------------------------