Issues with "make install"

[robotanarchy]

Hello mbed TLS developers,

I have experienced some issues with 'make install'.

• 1. Even when the no_test target was specified earlier, make install tries to build the test cases (because the install target depends on all):

make no_test
make install # tries to build test cases

• 2. At least with busybox as ln-implementation, make install will create symbolic link loops for all binaries. From the Makefile (line 36 in mbedtls-2.0):

 ln -sf $$f $(DESTDIR)/bin/$$o ; \

From what I understand, $$o is always empty?
(What happens is, that the binary gets overwritten by a symlink to itself!)

• 3. When compiling statically, only the libmbedtls.a file get copied. libmbedx509.a and libmbedcrypto.a are missing. When compiling against libmbedtls.a alone, lots of undefined reference errors pop up - because these functions are in the other .a files, so we definitely need them.

I've built mbedtls-2.0 on sabotage linux (my recipe with workarounds for these issues).

[mpg]

Hi, and thanks for your report!

I believe the third issue has been fixed in the meanwhile: 52a5079

Regarding the first issue, would you be happy with an option such as make install NOTEST=1 for this case?

Regarding second issue, this is a remainder of the time when we installed under both name (polarssl and mbedtls), I'll remove the ln line.

PS: which name would you like us to use for attribution in the ChangeLog?

[robotanarchy]

Hi mpg,
the make install NOTEST=1 sounds fine for me.

For attribution in the changelog, just use my username 👍

[mpg]

Hi, in the meantime I noticed we don't even need NOTEST=1, since the tests are never installed, so we can just have install depend on no_test. Simpler for everyone.