diff options
Diffstat (limited to 'example_avr_double/README.txt')
| -rw-r--r-- | example_avr_double/README.txt | 22 |
1 files changed, 22 insertions, 0 deletions
diff --git a/example_avr_double/README.txt b/example_avr_double/README.txt new file mode 100644 index 00000000..0090d729 --- /dev/null +++ b/example_avr_double/README.txt @@ -0,0 +1,22 @@ +Some processors/compilers, such as AVR-GCC, do not support the double +datatype. Instead, they have sizeof(double) == 4. Because protocol +binary format uses the double encoding directly, this causes trouble +if the protocol in .proto requires double fields. + +This directory contains a solution to this problem. It uses uint64_t +to store the raw wire values, because its size is correct on all +platforms. The file double_conversion.c provides functions that +convert these values to/from floats, without relying on compiler +support. + +To use this method, you need to make two modifications to your code: + +1) Change all 'double' fields into 'fixed64' in the .proto. + +2) Whenever writing to a 'double' field, use float_to_double(). + +3) Whenever reading a 'double' field, use double_to_float(). + +The conversion routines should be as accurate as the float datatype can +be. Furthermore, they should handle all special values (NaN, inf, denormalized +numbers) correctly. There are testcases in test_conversions.c. |