[Buildroot] openssl 1.1.x deprecated option

[Matthew Weber]

All,

On Wed, Feb 6, 2019 at 5:01 AM Peter Korsgaard <peter at korsgaard.com> wrote:
>
> >>>>> "Thomas" == Thomas Petazzoni <thomas.petazzoni at bootlin.com> writes:
>
>  > Hello,
>  > On Wed, 6 Feb 2019 04:03:09 -0600
>  > Matthew Weber <matthew.weber at rockwellcollins.com> wrote:
>
>  >> I was thinking about  how to manage the risk of a version bump vs
>  >> backport patches and found another possible solution.
>  >>
>  >> For openssl 1.1.x there are a series of deprecated APIs for items like
>  >> EVP_MD_CTX* which are now disabled and seem to result in 1/2 of the
>  >> failures. Would we entertain on some packages adding the libopenssl
>  >> "enable-deprecated"  configure option [1] so that it re-enables those
>  >> options (could do this like we currently do with a kconfig package = y
>  >> condition in the libopenssl.mk)?   Both mongodb and sqlcipher which
>  >> are currently failing should be resolved with this approach.
>  >>
>  >> Is this worth testing out / proposing?
>
>  > We could certainly have a BR2_PACKAGE_LIBOPENSSL_ENABLE_DEPRECATED
>  > option that enables those deprecated APIs, and have the packages that
>  > need that do:
>
>  >      select BR2_PACKAGE_LIBOPENSSL_ENABLE_DEPRECATED if BR2_PACKAGE_LIBOPENSSL
>
>  > Thanks to this option, a "git grep
>  > BR2_PACKAGE_LIBOPENSSL_ENABLE_DEPRECATED" allows to quickly identify
>  > which are the remaining packages that still need those deprecated APIs.
>
> Yes, that may be the most pragmatic option. Based on the statement in
> https://github.com/openssl/openssl/issues/4985 it does sound like we
> need to adjust the code (or CFLAGS) of each package using this option to
> add:
>
> #define OPENSSL_USE_DEPRECATED 1
>
> Before including any openssl headers, which isn't nice - But still
> better than complicated patches.
>

After some investigation the key item missing from the
enable-deprecated support is the HMAC_CTX and other memory allocation
compatibility functions which are now needed (they moved from stack
variables to heap allocations).  So not sure this idea holds water at
this point.

Matt