Skip site navigation (1)Skip section navigation (2)
Date:      Wed, 29 Oct 2014 15:47:25 -0700
From:      Garrett Cooper <yaneurabeya@gmail.com>
To:        Ian Lepore <ian@FreeBSD.org>
Cc:        Gyrd Thane Lange <gyrd-se@thanelange.no>, FreeBSD Current <freebsd-current@freebsd.org>, Mark Johnston <markj@freebsd.org>
Subject:   Re: buildkernel: make[2]: exec(ctfconvert) failed (No such file or directory)
Message-ID:  <4C533C42-2919-4E96-856C-FB6369F453FB@gmail.com>
In-Reply-To: <1414616599.17308.141.camel@revolution.hippie.lan>
References:  <20141028235011.543be3ea@onyx.thanelange.no> <1414537299.17308.28.camel@revolution.hippie.lan> <20141029003515.28e26444@onyx.thanelange.no> <CAGHfRMB5y_wwnvDkfr=p-A=a06rb1z74Am9rCk7jOy-_9X4Krg@mail.gmail.com> <20141029012432.41e22c7a@onyx.thanelange.no> <20141029143850.5af41378@onyx.thanelange.no> <1414593742.17308.72.camel@revolution.hippie.lan> <20141029212454.6fcfc3ac@onyx.thanelange.no> <1414616599.17308.141.camel@revolution.hippie.lan>

next in thread | previous in thread | raw e-mail | index | archive | help

> On Oct 29, 2014, at 14:03, Ian Lepore <ian@FreeBSD.org> wrote:

...

> It appears that src/Makefile.inc1 already does the right thing in
> regards to building ctconvert as a bootstrap tool when it needs to.

Yes and no. The assumption with the version checks is the the tool is presen=
t on the base system (which might be invalidated by build options or install=
 root pruning) and doesn't necessarily need to be rebuilt. It's an optimizat=
ion that works in most, but not all cases.

This is why make in /Makefile has testcases to ensure that it can bootstrap t=
he build properly.

>> happen to notice it sooner because I have no ctfconvert installed on
>> the host beforehand. Actually, mine is the better situation in that the
>> build stopped with an up front error instead of using the wrong
>> (possible outdated) tools and introduced other subtle errors.
>>=20
>> I believe this is reproducible on anybodys system by deleting/moving
>> the ctfconvert tool away from the host, before trying to build a dtrace
>> enabled kernel.
>=20
> I think you'll find that that is true of any number of programs
> in /usr/bin including, for example, cc or ar or ld.

Which gets invalidated periodically, ie you can't build current on certain s=
table versions.

>> I see elsewhere in the thread that Garret and Mark are onto the real
>> problem and solution.
>=20
> I don't think so.  Mark referred to a problem with the wrong version of
> ctfconvert, and he said he thought it had been fixed.  Your problem
> wasn't corrupted files due to wrong-version.
>=20
> We must not let this situation turn into some kind of mythology about an
> incompatibility that doesn't exist.  You reported that a tool was
> missing during the build, we know why it was missing, and we know what
> to do to make it not be missing anymore.  Problem solved.

In the simple case yes, but not in our case. Our case involves making fixes t=
o ctfconvert, without bumping the osreldate, because we've forked FreeBSD at=
 a particular point in time. ctfconvert is not compatible between the build h=
ost and the src checkout, so our ctf data gets corrupted when the host copy o=
f ctfconvert gets run on the binaries.

Cheers..=



Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?4C533C42-2919-4E96-856C-FB6369F453FB>