Get length of a string variable using the C interface?

[johnomotani]

To read a string variable using the C interface, I need (I think) to pre-allocate a char buffer. To do that correctly, I need to know how long the string is. How do I get the length of the string using the C interface? As I understand, a single string is not defined as an array, so I don't think the normal array-size functions help here?

[eisenhauer]

Hmm. This appears to be a feature gap in the C api. For many of the functions that return strings (like adios2_variable_type_string()), you call the function with a NULL buffer pointer first to get the size filled in, then call it again with a properly allocated buffer. This is not possible with the simple adios2_get() API that is used for string variables. The only reasonable workaround I can suggest is to use a buffer that you know is large enough. That's not a great answer, but it's the only thing I can think of at the moment.

[johnomotani]

That is unfortunate. I asked the question because I'm trying to fix a segfault due to a file I have where the string is longer than a buffer size hard-coded in a package (that's not mine!).

If I managed to write a PR for a new function like adios2_get_string_variable_size(), would that be welcome (I'm happy to give it a go!)? Would a feature request would be preferred?

[eisenhauer]

I've got a PR on this problem already in the queue. You're welcome to snag it in the interim to try.

[johnomotani]

Awesome, thanks!

[eisenhauer]

OK, PR #4801 that addressed this issue has been merged with master. It will appear in release 2.12 when that happens (not yet scheduled).

[johnomotani]

@eisenhauer is there the same issue for an attribute whose value is a string? I'm looking at adios2_attribute_data().

[johnomotani]

Sorry for not checking/noticing this the first time round!

[eisenhauer]

Oy. Yes, it seems that the same issue exists for attributes. I should have noticed that too. And unfortunately it's worse because because attributes can be an array of strings (for some reason that I'm sure seemed good at the time). Sigh. Let me see about extending the fix.

[eisenhauer]

OK, a 2nd PR (#4804) has been merged that addresses the problem with attributes.