I added some logging to the subversion export script to find which files were modified by the end-of-line normalization. The complete list is currently as follows:
nt-tools/init_nt nt-tools/tools/lib
These are a bit special in that they use LF line terminations, but contain literal CR characters on some lines as part of shell arguments. I think it's safest to simply remove the eol-style property on these particular files, and treat them as "binary".
packaging/fink/README packaging/fink/nettle.info packaging/fink/pike7.6.info
The newest versions of these files in 7.6 use CRLF line terminations, but I think that is an accident, as the earlier versions did not, and neither do those which are currently in 7.7. Comments from fink people?
packaging/windows/Copying.txt packaging/windows/finalizer.pike packaging/windows/nilsson-pike.iss packaging/windows/pike.iss
These are used to create Windows dists, potentially from trees checked out on Unix, so CRLF eol-style is probably best here. For the ISS files it might even be required. I'll change the property.
src/UnicodeData-ReadMe.txt src/UnicodeData.txt
The earliest versions of the Unicode database used CRLF line terminations, but everything since Unicode 3.1.0 was imported using LF line terminations. Normalizing should be safe, but to be historically accurate I could set CRLF eol-style on those old versions. It would mainly affect 7.0 and older which still have the old Unicode versions at HEAD.
src/modules/_Crypto/rijndael_cbc_d_m.txt src/modules/_Crypto/rijndael_cbc_e_m.txt src/modules/_Crypto/rijndael_ecb_d_m.txt src/modules/_Crypto/rijndael_ecb_e_m.txt src/modules/_Crypto/rijndael_ecb_iv.txt src/modules/_Crypto/rijndael_ecb_tbl.txt src/modules/_Crypto/rijndael_ecb_vk.txt src/modules/_Crypto/rijndael_ecb_vt.txt
These used to have CRLF line terminations, but were explicitly changed to LF in the last day of 2001 (commit by nilsson). The file rijndaeltest.pike which uses these files has the corresponding line termination hardcoded in, so native eol-style will not work here. I'll change it to CRLF eol-style for the old versions, and LF eol-style for the newer. This only affects 7.4.
src/post_modules/Nettle/byte-hashes.sha1 src/post_modules/Nettle/byte-messages.sha1
These have CRLF line terminations, but sha1test.pike which uses them strips all CR characters before processing, so either LF or CRLF terminations should work equally well. Unless someone protests, I'll keep native eol-style and perform the normalization for these files.
src/post_modules/COM/examples/ads.pike src/post_modules/COM/examples/shelltest.pike src/post_modules/COM/examples/word.pike src/post_modules/COM/examples/word2.pike
These have CRLF line terminations, presumably because they have been developed on Windows. Since they are pike scripts, the line termination chosen will not affect their function. So the question is whether there is a compelling reason to have CRLF terminations in these particular files even when checked out on Unix. Comments?