Hello!

If I compile a script containing string literal to bytecode, it compiles successfully. But when I try to load this bytecode in the application, it says for instance: "LoadByteCode failed. The bytecode is invalid. Number of bytes read from stream: 110". I'm using ScriptStdString implementation and the problem persists both in 2.28.2 and trunk (rev. 1933). Using string class itself and other add-on classes seems fine.

Test code (compiles and loads):
[source]int main() {

return 0;
}[/source]

Test code (compiles, but doesn't load):
[source]int main() {

}[/source]

UPD: if it matters, my system is x86 Linux, compiler: gcc (GCC) 4.9.0 20140507 (prerelease).

UPD2: windows 8 x64 with VS2013 shows the very same results.

A typical workflow I use to reproduce this behaviour:

Dump configuration with modifier asrun program. Compile sample code with this configuration into bytecode with separate compiler (modified asbuild), then try to run the compiled bytecode with asrun, which generated the configuration. Anything, containing string literals makes it fail to load the bytecode:

[source]

string s1 = "123";

string s2("abc");

s1 = "qwe";

writeln("Hello");

dictionary dict = {{"test", 42}};

[/source]

Modifications of asbuild are not very significant (hardcode array's template callback to allow its registration). In asrun i've removed building module from source with CScriptBuilder and added loading from bytecode like this:

[source]

[/source]

Error checks are removed for brevity. Maybe I'm doing something wrong or forgot something?

This file was generated with my version of asrun. Compilation was done fine, so i didn't think the problem could be in it.

Config parsing in asbuild is the same as in original asbuild sample, just with some workaround for template callbacks.

From what I see in writing/reading configuration file, engine properties and string factory are dumped and restored correctly and I don't see anything, that can affect bytecode generation. Both asbuild and asrun are linked against the same AngelScript library, so there should be no problem with version incompatibility.

After some experiments, I've found out, that without using string pool everything works fine. When using string pool, for some reason it outputs non-pool version of string factory to configuration. So compiler supposes that string factory returns string and runner expects from it to return const string &.