A quick look shows that this is always initialized in InsertMemorySizes, but in the 8086 and jvm platforms the function code is overridden, so it looks like except for these 2 cases.
Those two are always cross-compiled platforms. I checked, and my code works to identify the compiler. Thanks again for looking into this.
Same as my answer to Bart: my question is very, very particular: can it be relied upon in each and every case?
Should it be considered hackery or useful?
That is not because this particular example, but about any section in the code being accessible.
And the code is definitely public - through the external modifier - but the double underscore tells me I may not rely on it. Except maybe for pure assembler, which is the real use case.
And don't be confused with Opaquedata: that is simply a feature I added for external libraries.
I left out the above example in my patch request, probably for that reason, but its use looks just fine.
Hence I wrote: probably only answerable by compiler devs. I will ask on the development list just in case.