Daniel’s weekly report
January 28, 2022
Happened this week
-
The email I received, tweeted about and finally blogged about got some serious attention which distracted me somewhat in the start of the week. I tried to get a conversation going with them but I seem to have failed. The company name behind my blacked out fields in the published emails is MetLife.
-
The most likely never-ending troubles with host names ending with a trailing dot got another episode this week when I landed a change in curl that partly reverts a change done almost eight years ago which should make curl now handle these names more aligned with how current browsers use them. There is no doubt in my mind that we will have reasons to revisit this topic again in the future. Hopefully it will take another eight years before we do.
-
I’ve written an initial version of support for –json, and I think I’m going to land this for inclusion already in the next curl release. The discussion is still going on for the more advanced functionality to actually create JSON, and I haven’t yet decided on what I think about it or exactly how that is going to play out. I think we need more time to work on that and maybe write up a first attempt to let us try it out a little before we make up our minds, needing longer time than until the feature freeze next week. Feel free to jump in and share your views on this with us!
-
Another brain storming session was started with my asking for feedback about a new libcurl API for accessing response headers. It immediately showed that the easiest approach I had in mind probably was a little bit too easy and narrow-minded. I think maybe with some more thoughts and iterations this could lead to a good API that will save application authors’ time and energy in the future.
-
I did an online presentation for the Rust Linz meetup about “curl with rust”. The live-streamed version is available for watching, but I believe it will be made available in a more polished version later. I also made the slides available.
-
I got really fed up with NSS this Friday morning when I realized they have completely removed their online API documentation, making me ask the question Has the time come to drop NSS? (from being supported by curl) on the libcurl mailing list. They are clearly making it very hard for outsiders to use their library.
-
Reminder: get curl support for your company. This is what lets me work on curl this much.
Blog posts
Coming up
- curl roadmap 2022
- Working the header API proposal
- A presentation on “safe code” for a Swedish broadcasting company
- curl feature freeze
- closing in to the 1,000th curl commit author
Feedback
January 21, 2022
Happened this week
-
It was pointed out to me that the look if this log is not very phone friendly, which disappointed me. It is just a default GitHub theme and I thought they assured that… I haven’t changed theme yet since I don’t know which are good if any, so I haven’t decided on how to go forward with that.
-
Feedback is tricky. Thursday afternoon my time I posted an email to the curl-users mailing list and almost in parallel I asked people on Twitter the same question: Is it time to add some JSON aware support to the curl tool? It took off immediately on hackernews and reddit and the apparently intense interest made me also create a separate GitHub discussion for it. Then what? It smells of the good old bike shed problem: everyone can easily have an opinion on the matter. 1/3 are completely against the idea (layer violation, feature creep etc), 1/3 immediately pointed out numerous other things that apparently also must be supported for this to work and 1/3 seemed to mostly agree but … different tweaks. So yes, much more and thorough feedback than I am used to, but also so disparate and wide-ranging that it is really hard to maneuver through. My plan it to take it slowly, move forward with baby steps, go where I get the least amount of complaints and see where it takes us.
-
On the Security Now podcast episode 853 (starting about one hour in), the host Steve Gibson talks about last week’s URL confusion vulnerability research paper
-
A tsunami of web traffic on the curl site apparently started just before Jan 16. I have no idea what this is since I don’t log anything. Fastly just deals with it for us right now, but if the pattern sticks going forward I will need to investigate. (The image below is from the Fastly control panel. The Y axis and the averages, are amounts per hour. The X axis is the last week.)
-
I posted my blog post called Enforcing the pyramid explaining a little why it is hard to fix the problem of low quality and occasional security issues appearing in small but widely used open source projects. I think often the real reasons for the problems are ignored and people mostly repeat “lack of funding”. It might still be a lack of funding, but the problem is also where funding should go and how to manage that funding to actually enforce the software components others rely on. Everyone’s good intention are unfortunately not going to change a lot I think, for pure economic reasons: companies make a lot of money this way.
-
I discovered that curl still had source code left for Watcom compilers, even after we removed the support for building with this dead compiler well over a year ago so I went through and cleaned up. It made me realize that BeOS also had similar traces and I subsequently removed those as well. Less code is good. Especially when the removed code is never used anyway! Inspired by having found leftovers for these two legacy systems I wrote up a script that collects and analyzes all defines used in
#if
and#ifdef
lines in the product code in an effort to find more to clean off. The result shows that we use no less than 1,154 (!) different symbols in such lines, but I didn’t find any more equally obvious targets to clean up…
Blog posts
Coming up
- More curl roadmap 2022 thoughts
- More json thoughts and maybe code
- A header API proposal for libcurl
- I’ll do an online presentation called curl with rust for the Rust Linz Meetup on Thursday January 27th.
Feedback
January 14, 2022
Happened this week
-
The curl feature window is now open and we’ve started to merge outstanding pull-requests for new features. This also means that the next curl release is now going to be called 7.82.0.
-
The report on URL parser confusion that I have I mentioned before went public on Monday. It discusses problems and inconsistencies among URL parsers that can lead to vulnerabilities. I also joined Snyk’s live stream and talked about the report and the topic. I also wrote a separate blog post about it, see below.
-
Microsoft took a big step forward this week and shipped an upgrade to curl for Windows. They had stalled on version 7.55.1 for years and now they took a giant leap in one go straight up to 7.79.1. Dustin Howett said they intend to keep up-to-date better going forward.
-
In Microsoft’s upgrade note, they specifically highlight that they’ve addressed CVE-2021-22947 but they also confusingly call that problem a “Remote Code Execution Vulnerability” - which it isn’t. After some back-channel talk, it has been explained that they fixed a whole range of problems with that upgrade and it includes one RCE. This seems reasonable, as we list quite a few known vulnerabilities in 7.55.1. 35 to be exact.
-
At wolfSSL we have started the year by internally discussing the curl 2022 Roadmap or more specifically what I will work on in curl this year. I took the subject to the mailing lists and I also got a lot of feedback on my Twitter question for lessons from curl alternatives. We are going to customers starting next week and I will most likely do a webinar on the topic maybe early February.
-
I did an internal curl presentation at wolfSSL since we have quite a few new faces at the company over the last year. What’s curl, what does it do, existing customers, how do we (want to) sell curl and something about the possible roadmap 2022. In the big wolfSSL family of libraries and products, curl is certainly the little brother business-wise. Room to grow.
-
Our friends in the quiche project landed a changed that brought the
quiche_conn_peer_cert()
function that would allow us to verify the server certificate when doing QUIC connections (for HTTP/3) with this library. Ironically though, it still turned out inadequate for curl’s purposes and we instead took a different route when Alessandro Ghedini stepped up and took us a lot closer to the goal. Without using that new function. -
Vaxxed v3.
Blog posts
Coming up
- More PR review and merging.
- A libssh2 release is due and it might happen this coming week
- curl roadmap 2022
- A blog post around the OSS pyramid
Feedback
January 7, 2022
Happened this week
-
Because I had some time off during Xmas I didn’t do any weekly report last week and therefore this edition might include stuff I did last week when I was supposed to not do much.
-
On January 5 we released curl 7.81.0, and I got to be the release shepherd for the 205th curl release ever. There have been some issues reported on this version, in particular some annoying build related quirks, but so far none of them have been deemed serious enough to warrant any follow-up patch release. It looks like we can open the feature windows next week and plan for a 7.82.0 release for March 2 - precisely according to the schedule.
-
I have brought up the discussion about “curl roadmap 2022” on the mailing lists as well as on Twitter in order to listen in on what people are thinking about and want for curl these days. I hope to put together a 2022 roadmap that at least outlines what I plan to work on for curl this year. It is also intended to highlight the fact that paying customers/companies of mine can affect this. Get in touch!
-
Almost every single downstream curl packager has its own set of custom patches applied. Many of those changes are just patches that affect particular details in the build and are not details we would like to modify in the project’s upstream code. After I did a closer inspection of the patches used by Debian, FreeBSD and msys it is also clear that there are functional changes all over that the downstreams haven’t submitted to us. In most cases they also ship and get applied without much explanations or documentation making it hard for us to do anything about them. I assume the patches are or were better motivated at some point.
-
I have slowly started to move CI jobs away from Zuul CI over to other services. When Travis “went rogue” last summer, we moved almost all of those CI jobs over to Zuul CI and we were happy having managed the shutdown in such an elegant and swift manner. At the most, we had 31 different CI jobs running on Zuul (out of the about 100 jobs we run in total). Over time however, it has turned out that Zuul’s integration with GitHub is so flawed and that builds frequently don’t even show up in Zuul’s UI that I have felt forced to decide that they are not a preferred service either. It’s not any alarming problem but I plan to pick a few jobs every now and then and get them moved over to run on other services instead. Feel free to help me out with this.
-
Let me tell you about my hack of the week to increase my productivity: I’ve now created a build script that replaces my alias (
b
) I’ve been using for building curl. The previous alias would domake -C $HOME/src/curl -sj7
just as a convenient shortcut (it uses-s7
only to make my video streaming performance get less affected when I do live-coding sessions). The replacement now has an additional tweak: it also scans modified files to verify that the copyright year ranges are correct. curl’s copyright scan script is otherwise too slow to run against all files in every build command. The complete script looks like this:
#!/bin/sh
cd /home/daniel/src/curl
files=$(git diff origin/master --name-only)
if test -n "$files"; then
./scripts/copyright.pl $files
res=$?
if test $res != 0; then
exit 2
fi
fi
make -sj7
Blog posts
Coming up
- The curl feature window opens on Monday the 10th. There are plenty of pull- requests already waiting to get merged once that happens.
- There will be a report published on newly discovered URL problems. The report comes from people at Snyk and Claroty but I’ve read it and given it my thumbs up. I’ll do my own blog post about it too.
- There will be a live-streamed discussion about the above mentioned URL report that happens on Tuesday 15:00 UTC on Twitch where I will participate.
- I will continue to improve the HTTP/3 support. I want to get to fixing cert verification for curl + quiche. If our friends in the quiche project delivers, this might be a fine topic for me to also do a live-streamed session with. Follow me on Twitter or Twitch if you don’t want to miss it.