Update ExprTk

[ArashPartow]

1. Update ExprTk to 0.0.3 Release notes: https://www.partow.net/programming/exprtk/exprtk_release_notes_v0.0.3.txt
2. Adaptor update
3. Minor fixes and cleanups in exprtk, scalar, computed functions

[ArashPartow]

@texodus Looks like a couple of adaptor methods and free functions need to be added to perspective::t_tscalar:

1. to_uint64
2. cast operator to uint64 from type

I'm ok with adding them, would you want it all in the same PR or two PRs, initially the additional adaptor methods and then this PR?

[texodus]

One PR would be ideal.

[texodus]

I'm not sure if you can see our CI logs externally, but this PR does not compile for different reasons on every architecture we support.

I have not had time to test this on anything but my development machine yet, but I've rebased your PR and fixed the local compilation errors here, which you can cherry pick.

[ArashPartow]

Finally got all the tests passing. Minor issues some of the test results, which revealed a parsing issue in ExprTk.

That being said there is a change I couldn't resolve:

-    struct random : public exprtk::igeneric_function<t_tscalar> {
+    struct random final : public exprtk::igeneric_function<t_tscalar> {
         random();
-        ~random();
+       ~random();

-        t_tscalar operator()(t_parameter_list parameters);
+        t_tscalar operator()(t_parameter_list parameters) override;

         // faster unit lookups, since we are calling this lookup in a tight
         // loop.
-        static std::default_random_engine RANDOM_ENGINE;
-        static std::uniform_real_distribution<double> DISTRIBUTION;
+        std::mt19937 generator;
+        std::uniform_real_distribution<double> distribution{ 0.0, 1.0 };
     };
 -// Set up random number generator
-std::default_random_engine random::RANDOM_ENGINE = std::default_random_engine();
-std::uniform_real_distribution<double> random::DISTRIBUTION =
-    std::uniform_real_distribution<double>(0, 1);
-
-random::random() : exprtk::igeneric_function<t_tscalar>("Z") {}
+random::random()
+: exprtk::igeneric_function<t_tscalar>("Z")
+{
+    std::random_device device;
+    std::array<unsigned int,std::mt19937::state_size> seed;
+    std::generate_n(seed.data(), seed.size(), std::ref(device));
+    std::seed_seq seq(std::begin(seed), std::end(seed));
+    generator.seed(seq);
+}

It seems the use of std::random_device in a WASM/NODERAWFS context is the issue. The implementation of std::random_device is typically done by using /dev/random or /dev/urandom (at least in a posix context). However if those devices are not available, this leads to UB which resulted in spurious crashes during the js tests. I tried several approaches to getting it to work, however was not successful.

The main reason for this particular change was to properly seed the engine, as the way it is currently being setup is not correct. It will use an internal state based on whatever is the memory contents of the allocation. Which could be random or all zeros etc.