Re: Problems with CURL_DOES_CONVERSIONS/HAVE_ICONV build on EBCDIC platform
On Wed, 11 Apr 2018, Stephan Mühlstrasser wrote: The easiest way to get a working library on EBCDIC again would be to conditionally compile the old macros that were used before the introduction of curl_ctype.c, as you also mentioned in one comment in pull request #2269. This worked well for us before. Would that be an acceptable solution? If that's fine by you, it's fine by me! =) You're the one running and using this setup so your advice and recommended path forward will carry a lot of weight. I figure restoring functionality should be prio one. I felt a need to "fix" the ctype stuff for curl when I ran into troubles with isalnum() being locale-dependent. Described here: https://daniel.haxx.se/blog/2018/01/30/isalnum-is-not-my-friend/ -- / daniel.haxx.se--- Unsubscribe: https://cool.haxx.se/list/listinfo/curl-library Etiquette: https://curl.haxx.se/mail/etiquette.html
Re: Problems with CURL_DOES_CONVERSIONS/HAVE_ICONV build on EBCDIC platform
Am 11.04.18 um 14:32 schrieb Daniel Stenberg: We look at every single filed bug report and pull request. (Sometimes we're slow to act on them though, as we're not that many developers who work on random bugs and PRs.) The world is big and diverse and we have a lot of users who are using systems or curl features that a large portion of the world does not. When you experience bugs on a niche platforms or with features not used widely, there's a chance that A) you're alone to report and experience the bug and even B) there's no regular developer around who's using it. I'm convinced curl on EBCDIC is one such niche with few devlopers. We don't get many bug reports about EBCDIC, which I believe is because we don't have many users using EBCDIC... We do have a travis build making sure CURL_DOES_CONVERSIONS builds work and are functional, but that's not using EBCDIC. Not enough test and CI infrastructure for something as impactful as EBCDIC will quite naturally lead to bugs getting introduced over time. Thank you, Daniel. It's clear to me that EBCDIC support is a niche and that I need to become active myself for getting things fixed, but it's good to know that bug reports about EBCDIC problems are heard. 1) subjectAltName check for HTTPS connections In lib/vtls/openssl.c in function verifyhost() two strings with different encodings are passed to Curl_cert_hostcheck(), causing the check to always fail. I have a potential fix for this. Please file a PR! Will do. 2) ASCII-based character classification > Exactly, but as the comment says: it's next to impossible to work on this and get it right without actually being able to test and verify it on such a system. Please file a PR for a fix that makes it work for you! I will try to come up with a pull request as well, but I need some mentoring for this. The easiest way to get a working library on EBCDIC again would be to conditionally compile the old macros that were used before the introduction of curl_ctype.c, as you also mentioned in one comment in pull request #2269. This worked well for us before. Would that be an acceptable solution? If those are the only problems, I consider things to be in a pretty decent shape still! That is certainly true, and it's amazing how well curl works on EBCDIC platforms in general! -- Stephan --- Unsubscribe: https://cool.haxx.se/list/listinfo/curl-library Etiquette: https://curl.haxx.se/mail/etiquette.html
Re: Problems with CURL_DOES_CONVERSIONS/HAVE_ICONV build on EBCDIC platform
On Wed, 11 Apr 2018, Stephan Mühlstrasser wrote: So my question is, what is the state of curl support for platforms that require CURL_DOES_CONVERSIONS and HAVE_ICONV? Are corresponding bug reports looked at and are pull requests wanted? We look at every single filed bug report and pull request. (Sometimes we're slow to act on them though, as we're not that many developers who work on random bugs and PRs.) The world is big and diverse and we have a lot of users who are using systems or curl features that a large portion of the world does not. When you experience bugs on a niche platforms or with features not used widely, there's a chance that A) you're alone to report and experience the bug and even B) there's no regular developer around who's using it. I'm convinced curl on EBCDIC is one such niche with few devlopers. We don't get many bug reports about EBCDIC, which I believe is because we don't have many users using EBCDIC... We do have a travis build making sure CURL_DOES_CONVERSIONS builds work and are functional, but that's not using EBCDIC. Not enough test and CI infrastructure for something as impactful as EBCDIC will quite naturally lead to bugs getting introduced over time. 1) subjectAltName check for HTTPS connections In lib/vtls/openssl.c in function verifyhost() two strings with different encodings are passed to Curl_cert_hostcheck(), causing the check to always fail. I have a potential fix for this. Please file a PR! 2) ASCII-based character classification The functions in curl_ctype.c work for ASCII only. After checking the history if this file I found the following related GitHub pull request: ... So this apparently is a known problem for EBCDIC. Exactly, but as the comment says: it's next to impossible to work on this and get it right without actually being able to test and verify it on such a system. Please file a PR for a fix that makes it work for you! If those are the only problems, I consider things to be in a pretty decent shape still! -- / daniel.haxx.se--- Unsubscribe: https://cool.haxx.se/list/listinfo/curl-library Etiquette: https://curl.haxx.se/mail/etiquette.html
