nightly/PRs: IAR build errors with an internal error

[0xc0170]

Description of defect

This currently affects PRs and it is also in the latest nightly (was not in the day before). Something has changed in CI or example?

This is the example failing mbed-os-example-socket-stats - cc @ARMmbed/mbed-os-core

[ERROR] Internal error: [Front end]: assertion failed at: "/home/jenkins_ci01/slave_top/workspace/ArmReleaseBuild/core/translator/compiler_core/src/parser/edg/il.c", line 23626


Fatal error detected, aborting.

[mbed] Working path "/builds/ws/mbed-os-ci_build-IAR@3/examples/mbed-os-example-socket-stats" (program)

As nightly and also PRs (one is using master and one release branch), this does not indicate it is in the example itself but rather a change in CI within the last 15 hours or so?

cc @ARMmbed/mbed-os-test

Target(s) affected by this defect ?

K64F

Toolchain(s) (name and version) displaying this defect ?

IAR

What version of Mbed-os are you using (tag or sha) ?

mbed-os-99.99.99

What version(s) of tools are you using. List all that apply (E.g. mbed-cli)

CI related

How is this defect reproduced ?

I am testing locally , will update

[0xc0170]

I am able to reproduce it on example master

Scan: mbed-os-example-socket-stats
Compile [  0.1%]: mbed_tz_context.c
Compile [  0.2%]: MCR20Drv.c
Compile [  0.3%]: at24mac.cpp
Compile [  0.4%]: main.cpp
[ERROR] Internal error: [Front end]: assertion failed at: "C:\j\workspace\Arm832Build\core\translator\compiler_core\src\parser\edg\il.c", line 23630

If I revert the last commit: ARMmbed/mbed-os-example-socket-stats@22cdd95 , it builds

@evedon Something triggers IAR internal fault in the latest commit there.

[0xc0170]

I found the offending line:

mbed_stats_socket_t stats[MBED_CONF_NSAPI_SOCKET_STATS_MAX_COUNT] = {0}; If I remove {0}, no internal fault.

cc @kjbracey-arm

[0xc0170]

Found a fix, not ideal but will do the same functionality just avoiding the IAR error

[0xc0170]

ARMmbed/mbed-os-example-socket-stats#47 should fix it for now

[kjbracey]

I think the revised code is better - it's a big enough object to be worth not putting on the stack.

[0xc0170]

I also reported this to IAR via website support.