Re: Recent BoringSSL build breakage

[Willy Tarreau]

On Thu, Aug 01, 2019 at 11:17:36AM +0200, Emmanuel Hocdet wrote:
> Yep, i already built with the change.  Fix included.

Now pushed, thank you! Let's hope all checks turn green now.

> I'm looking at BORINGSSL_API_VERSION for compatibility evolution, but for now
> is not incremented as i would expect.

I think it's only for major changes but not minor one. Ideally they
should start to use "#define foo foo" when they define something so that
it becomes possible for their users to do stuff like the following :

   #if defined(BORINGSSL) && !defined(foo)
   static inline foo()
   { 
blah();
   }
   #endif

Cheers,
Willy

Re: Recent BoringSSL build breakage

[Emmanuel Hocdet]

Hi Willy,Le 1 août 2019 à 10:07, Willy Tarreau  a écrit :Hi Manu,On Travis CI there was a fairly recent regression on BoringSSL whichhappened between 03e09f3 and a7a0f99 a day ago. It breaks on definitionof EVP_PKEY_base_id() in openssl-compat.h, which was not modified, andI guess this issue was hidden by the simultaneous breakage of libresslby latest changes.It looks like latest BoringSSL now defines this function and that thedefinition above could be removed. Could you please have a look whenyou have a moment and possibly propose a patch so that we can fix thosebuild reports (especially if you find that other places need to betouched) ?For reference, first breakage :    https://travis-ci.com/haproxy/haproxy/builds/121281529Last known good build:    https://travis-ci.com/haproxy/haproxy/builds/121258130Yep, i already built with the change.  Fix included.I'm looking at BORINGSSL_API_VERSION for compatibility evolution, but for nowis not incremented as i would expect.// BORINGSSL_API_VERSION is a positive integer that increments as BoringSSL// changes over time. The value itself is not meaningful. It will be incremented// whenever is convenient to coordinate an API change with consumers. This will// not denote any special point in development. A consumer may use this symbol in the preprocessor to temporarily build// against multiple revisions of BoringSSL at the same time. It is not// recommended to do so for longer than is necessary.#define BORINGSSL_API_VERSION 9++Manu

0001-BUILD-ssl-BoringSSL-add-EVP_PKEY_base_id.patch
Description: Binary data

Recent BoringSSL build breakage

[Willy Tarreau]

Hi Manu,

On Travis CI there was a fairly recent regression on BoringSSL which
happened between 03e09f3 and a7a0f99 a day ago. It breaks on definition
of EVP_PKEY_base_id() in openssl-compat.h, which was not modified, and
I guess this issue was hidden by the simultaneous breakage of libressl
by latest changes.

It looks like latest BoringSSL now defines this function and that the
definition above could be removed. Could you please have a look when
you have a moment and possibly propose a patch so that we can fix those
build reports (especially if you find that other places need to be
touched) ?

For reference, first breakage :
https://travis-ci.com/haproxy/haproxy/builds/121281529

Last known good build:
https://travis-ci.com/haproxy/haproxy/builds/121258130

Thanks,
Willy