David Heffernan

ARM has to win in the end at least for portable devices.

This is cracked software, right?

There are commercial tools to support this. I sure websearch will find them. As soon as you link to anything that isn't delphi then the desire for a tool for Delphi is a bit pointless.

Although the real problem is the language design that makes this order affect the meaning of the program.

OK. Well I'm glad this thread has been useful!!!

I've no idea what your requirement is and what the problem is. You don't seem to understabd how floats work. I don't think this is going to be productive.

K and n are integers. All representable floats are of the form k*2^n. So 1/3, 1/10 aren't of that for, and aren't representable. I doubt that you need to store these numbers exactly. But we don't know your requirements.

You can see that it's impossible to store irrational numbers in floats. Indeed even rationals can't be stored exactly in floats, unless they can be expressed as above, k*2^n.

Yes but it's hard to see the relevance. Representable numbers have the form k*2^n. Perhaps you don't actually need exact representability. We don't know what your requirements are.

I don't know what this means

Of course the other problem with packing a record that should be aligned is that gives it alignment of 1 so they can be misaligned. These days that doesn't usually matter for x86/x64 but it can be a issue for some SSE2 instructions and for arm processors.

Even rational numbers aren't exactly representable. Examples include 1/3 and 1/10.

You mean the request to store pi exactly in a float ?

Hard to comment on this without seeing you actual code. This is all about the detail and we have none of it. For all I know you are using packed records and manually inserting padding. In which case, if that's what you've decided to do then you know how to fix it.

It would have been useful to say something like this, like you may need manual alignment if you need to use ancient compilers like Delphi 6. But even then, in most cases, aligned records work as expected. It's only in some very rare cases that this is not the case. Perhaps you have an example for us. Or has that knowledge been lost in the passage of time. For people that don't need to support 20 year old compilers, feel free to use the compiler to do the alignment.

Show us an example, with modern Delphi. For sure some ancient versions didn't get alignment correct. That's no longer the case. Compiler alignment always works for me, and I do a lot of it. And converting to 64 bit was trivial when I had to do that.

In your Delphi code you should be using {$MINENUMSIZE 4} at the point where you define the enums. An example from my own codebase: //Dragon4 {$MINENUMSIZE 4} type TDragon4DigitMode = (DigitMode_Unique, DigitMode_Exact); TDragon4CutoffMode = (CutoffMode_TotalLength, CutoffMode_FractionLength); {$MINENUMSIZE 1} I would however say that using offset methods to compare offsets is very good advice. But I definitely do not advise manual alignment. The compiler can do it perfectly well. You may as well get it to do that. Then, come the day you move to 64 bit, you won't have to change a thing.

This is bad advice. If you are using aligned records in Delphi, then the compiler uses the same alignment rules as for aligned C++ structs, and for aligned C# structs.

The things you are discussing now seem unrelated to storing pi in a float. I don't feel as though we understand the problem domain yet.

You can't store pi exactly in a float. What is the motivation for the question?

A dfm file is a source file

Definitely not impossible to find the error. Search for TDBLookupComboBox in all the .dfm and .pas files that are compiled.

This isn't true. Your class function calls the default constructor and that performs memory allocation. Free must be called. Instances of classes are heap allocated.

What have you tried so far?

What specifically do you mean by "help me translate"