[mgtremaine]

The openssl in SGUG works fine. openssl-1.1.1d-4.sgugbeta.mips I had assumed it was broken because openssh 8 could not read the keys. However that has been patched and I verified both the openssh server and curl and wget against openssl 1.1.1d do the right thing. Doesn't help with the MIPSpro build but you might check the patch files on that.

-Mike

[Raion]

The patch files on that are geared towards GCC. I've already tried applying those patch files but it does not fix the problem.

Generally speaking their patches are not very useful; it depends on the program, but the main issue with using GCC is that the patches are using their own patched headers/libdicl, and while GCC works fine for a lot of programs, SSL and crypto library support is one of the few areas this actually makes sense to try to squeeze a bit more performance from.

I'm not going to sit here and argue about whether or not there are merits to not using MIPSPro, but my testing on basic programs moreorless ruled out that GCC 4.7 and 8.2 have any speed differences, at least in those instances. Sometimes my hand compiled 5 or 7 did a tiny bit better than the 4.7.4 tardist floating around.

My focus is on MIPSPro, but since I don't have a GitHub account and refuse to be part of that organization, I can't file bug reports with Openssl, so I'm moving along to something else.

[mgtremaine]

Not looking to argue with you at all. Was mostly reporting that there is a working one. I did say (although saw i typo'd it) that is doesn't;t help with the MIPSpro issue. I think the person to ask is Canavan. His tardists where always clean. I see that he was working on it.

https://mta.openssl.org/pipermail/openss...00444.html

And then reply email probably works.

-Mike

[Raion]

I didn't think you were trying to debate with me directly, I thought I'd head off any discussion on whether or not MIPSPro is worth pursuing; gods know we have had enough. My policy is if the RSE produces a patch that works for my use case, I'll use it and credit the person who committed it there.

Unfortunately openssl's mailing lists for bug reports and development are no longer in use:

https://www.openssl.org/community/mailinglists.html

Canavan is here, actually on irixnet. His Tardists are unfortunately only for the 1.0.2s version, which moreorless corresponds with the breaks in openssl at 1.1.1, I never got 1.1.0 to work on IRIX in any version.

I'm currently working on seeing if I can get a libressl working but the makefiles are a pain. There's other ssl libraries I need to try, but yeah ideally someone would be able to open a GitHub issue on my behalf or something. I don't know why they deprecated their mailing lists...

[Raion]

https://github.com/openssl/openssl/issues/12456

Thanks to LadySerenaKitty for filing the bug for me. I sincerely hope that they decide to help take action on this but in the meantime drmadison passed me an Mbedtls patch that works, so I'll be using that to get Mbedtls working with curl.

I do want it to be stated for the record that yes, I make things harder for a relatively crappy developer like myself using MIPSPro with new software. Why do I try anyways? Because I have seen measured performance improvements and I think it's important to try anyways. If I don't start carving the path who is going to? This is my way of trying to put my money where my mouth is. Because of the small size of the Nekoware II team I'm making decisions I'm not really comfortable making after previous disasters, but it's something I have to do in absence of anyone else.

[Raion]

Looks like OpenSSL 1.1.1h is now fixed, I'm building it currently and will try a curl binary with it to see!

[Raion]

Code:

takemikazuchi 446# /opt/xeno/test/bin/curl -OLk https://irixnet.org/index.html
  % Total    % Received % Xferd  Average Speed   Time    Time     Time  Current
                                 Dload  Upload   Total   Spent    Left  Speed
100  1412    0  1412    0     0   3268      0 --:--:-- --:--:-- --:--:--  3276
takemikazuchi 447# file index.html
index.html:     ascii text
takemikazuchi 448# head -10 index.html
<!DOCTYPE html>
<html>
  <head>
    <meta charset="UTF-8">
    <title>IRIX Network - Home</title>
    <link href="assets/pages2.css" rel="stylesheet" type="text/css" media="all">
  </head>
  <body>
    <h1 id="nameleft"><span style="color:rgb(158, 0, 211);">IRIX</span>Network</h1>
    <p id="topbar">
takemikazuchi 449#

Think that settles it!

[Raion]

It looks like this regression is back with 1.1.1k

I am currently checking i and j versions to see where it was introduced.

[Raion]

Hmm, got some interesting results. Checking to see if some -OPT limits can fix this.

[Raion]

Yeah, so I figured it out. Optimizer bug. Have to use -O1 or set the -OPT flags so conservative it's not worth it. This appears to be a new regression.