Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Show HN: A fast, dependency-free traceroute implementation in pure C

github.com

Post reply on HN
Print view View original

21–30 of 34 posts

Re: Show HN: A fast, dependency-free traceroute implementation in pure C

#21

Post by 9029 »

raphman wrote: post #9

Honest question: why would this code clamp the reported round-trip time? By default, min = 0.05 ms and max = 800 ms [1]. if (rtt config.max_rtt) rtt = config.max_rtt; Wouldn't this hide bugs in the code or network anomalies? Replies from localhost seem to typically arrive in less than 50 µs. Comments in an earlier version [2] make no sense to me: /* Use standard timersub for more accurate results */ if (rtt 1000) rtt…

It has now been changed to

  if (rtt
https://github.com/davidesantangelo/fastrace/blob/e8b19407a4...

Re: Show HN: A fast, dependency-free traceroute implementation in pure C

#22

Post by johnisgood »

matltc wrote: post #18
johnisgood wrote: post #15

Earlier quoted context omitted.

We live in a world where now using "-Wall -Wextra" is a positive outlier. :D God damn. I have ALWAYS used these options, along with "-pedantic", "-std=c99" and so forth.

I picked up C for fun last year and this is exactly the flags I have always used by default. Can't remember where I picked that up, but glad to hear I'm doing it right

Yes you are.

I always use "-std=c99 -Wall -Wextra -Wpedantic -Werror". You could replace "-Wpedantic" with "-pedantic" though (it is more supported). You may omit "-Werror".

Sometimes I also use "-D_XOPEN_SOURCE=700" and "-D_FORTIFY_SOURCE=2" along with "-fstack-protector-strong".

For debug builds you want "-O0 -g" at the very least.

I also have a make target that uses "scan-build", "cppcheck", and "clang-tidy".

Re: Show HN: A fast, dependency-free traceroute implementation in pure C

#23

Post by anonymous908213 »

9029 wrote: post #21
raphman wrote: post #9

Honest question: why would this code clamp the reported round-trip time? By default, min = 0.05 ms and max = 800 ms [1]. if (rtt config.max_rtt) rtt = config.max_rtt; Wouldn't this hide bugs in the code or network anomalies? Replies from localhost seem to typically arrive in less than 50 µs. Comments in an earlier version [2] make no sense to me: /* Use standard timersub for more accurate results */ if (rtt 1000) rtt…

It has now been changed to if (rtt https://github.com/davidesantangelo/fastrace/blob/e8b19407a4...

And the update message has a reference to "50µs localhost responses", indicating the comment calling the code out was directly fed into a prompt:

"Fixed Removed artificial RTT clamping that was hiding legitimate network measurements Previously clamped RTT between 0.05ms and 800ms Now reports actual values including sub-50µs localhost responses and >800ms satellite/long-distance links Added sanity check for negative RTT to detect clock issues without corrupting data This fix restores full diagnostic capability for detecting network anomalies like bufferbloat and measuring true round-trip times across all network types."

It shouldn't be legal to vibe this hard, honestly. If convicted in court, you should face punishment of, say, XXX hours doing something actually useful to society with your own two hands.

Re: Show HN: A fast, dependency-free traceroute implementation in pure C

#24

Post by 9029 »

johnisgood wrote: post #22
matltc wrote: post #18

Earlier quoted context omitted.

I picked up C for fun last year and this is exactly the flags I have always used by default. Can't remember where I picked that up, but glad to hear I'm doing it right

Yes you are. I always use "-std=c99 -Wall -Wextra -Wpedantic -Werror". You could replace "-Wpedantic" with "-pedantic" though (it is more supported). You may omit "-Werror". Sometimes I also use "-D_XOPEN_SOURCE=700" and "-D_FORTIFY_SOURCE=2" along with "-fstack-protector-strong". For debug builds you want "-O0 -g" at the very least. I also have a make target that uses "scan-build", "cppcheck", and "clang-tidy".

While we are at it, here are some more useful warning flags I have used: https://github.com/cpp-best-practices/cppbestpractices/blob/.... Some C++-only though, some are a bit opinionated (like -Wsign-conversion) and some useful C-only flags might be missing.

Few C-specific references I found just now, but haven't tried myself yet:

https://github.com/systemd/systemd/blob/0885e4a6e7ca93d3aef8... https://github.com/airbus-seclab/c-compiler-security

Also a good idea to regularly run the program with sanitizers, using them in tests is a good way to do that I think. Why not during development as well if the performance is acceptable for that specific program.

Re: Show HN: A fast, dependency-free traceroute implementation in pure C

#25

Post by jstimpfle »

pjmlp wrote: post #13
pavlov wrote: post #6

It used to be that if someone released a tool that's 700 lines of C, they probably had an actual need and a problem they solved by writing that code because debugging even that amount of C tends to be non-trivial. Today all bets are off. Does the tool do anything anybody needed? Does it work? Who knows. It might just be 700 lines of convincing-looking C churned out by a model.

Well I was surprised that it actually makes use of "-Wall -Wextra" as good practice. However both PVS-Studio and clang-tidy have a few complaints about the code, since it is a single file, it is rather easy to try out on Compiler Explorer. https://godbolt.org/z/n4M1vGccq As for your remark, most folks seem to have not followed that C authors also created lint in 1979, Dennis Ritchie proposed fat pointers to WG14, Pla…

I've looked at a couple of these complaints by clang-tidy, and as it unfortunately often is, all of them were false positives and overzealous nitpicks. All the complaints about memcpy and memset for example, clang-tidy could easily be improved to see that these are just fine and being dogmatic about using the "new and right way" to do things is not helpful.

In practice I've found -Wall with GCC to offer a good warning level and clang-tidy to not offer a lot of constructive feedback (besides it being very slow). For more ambitious projects, it's possible to fine-tune GCC warnings.

You can also, you know, just _use_ a program and see if there are any anomalies when running it. With some discipline to code structure, many problems get hit on the first run, and extensive testing can come a lot closer to static verification than you would think. For non-real-time constrained stuff there is also valgrind and other run-time instrumentation.

Re: Show HN: A fast, dependency-free traceroute implementation in pure C

#26

Post by pjmlp »

jstimpfle wrote: post #25
pjmlp wrote: post #13

Earlier quoted context omitted.

Well I was surprised that it actually makes use of "-Wall -Wextra" as good practice. However both PVS-Studio and clang-tidy have a few complaints about the code, since it is a single file, it is rather easy to try out on Compiler Explorer. https://godbolt.org/z/n4M1vGccq As for your remark, most folks seem to have not followed that C authors also created lint in 1979, Dennis Ritchie proposed fat pointers to WG14, Pla…

I've looked at a couple of these complaints by clang-tidy, and as it unfortunately often is, all of them were false positives and overzealous nitpicks. All the complaints about memcpy and memset for example, clang-tidy could easily be improved to see that these are just fine and being dogmatic about using the "new and right way" to do things is not helpful. In practice I've found -Wall with GCC to offer a good warnin…

Instead of ranting, you should have realized that is the default output without configuration file, which isn't that easily to provide in compiler explorer, without going through the trouble of a project template.

Naturally on a real project there would be an heavily customised static analysis tool, that would only allow a build to succeed with the feedback from the SecDevOps team, alongside feedback loop from pentesters.

We have seen how far just _use_ the program has been a thing tracking down C security issues for the last 37 years, starting with Morris Worm.

And to quote Dennis Ritchie,

> To encourage people to pay more attention to the official language rules, to detect legal but suspicious constructions, and to help find interface mismatches undetectable with simple mechanisms for separate compilation, Steve Johnson adapted his pcc compiler to produce lint [Johnson 79b], which scanned a set of files and remarked on dubious constructions.

-- https://www.nokia.com/bell-labs/about/dennis-m-ritchie/chist...

Re: Show HN: A fast, dependency-free traceroute implementation in pure C

#27

Post by checker659 »

anonymous908213 wrote: post #17
checker659 wrote: post #10

Earlier quoted context omitted.

It could be a vehicle to learn about traceroute / ICMP.

For whom? What is the creator going to learn from pasting code they've never read? What are readers going to learn from reading code the "author" themselves didn't read, let alone write? If you want to learn about something, reading an LLM-generated repo seems to be about the worst possible way to do it. That's not even to say LLMs are useless for learning; you could ask directly about concepts without having it writ…

I meant that IFF the author wrote it themself. IFF.

Re: Show HN: A fast, dependency-free traceroute implementation in pure C

#28

Post by anonymousiam »

I just built and ran it.

Unlike traceroute and mtr, this utility must be run as root.

fastrace 1.1.1.1

fastrace 0.2.1

Tracing route to 1.1.1.1 (1.1.1.1)

Maximum hops: 30, Probes per hop: 3, Protocol: UDP

TTL │ IP Address (RTT ms) Hostname

────┼───────────────────────────────────────────

Error creating ICMP socket. Are you running as root?: Operation not permitted

Re: Show HN: A fast, dependency-free traceroute implementation in pure C

#29

Post by jstimpfle »

pjmlp wrote: post #26
jstimpfle wrote: post #25

Earlier quoted context omitted.

I've looked at a couple of these complaints by clang-tidy, and as it unfortunately often is, all of them were false positives and overzealous nitpicks. All the complaints about memcpy and memset for example, clang-tidy could easily be improved to see that these are just fine and being dogmatic about using the "new and right way" to do things is not helpful. In practice I've found -Wall with GCC to offer a good warnin…

Instead of ranting, you should have realized that is the default output without configuration file, which isn't that easily to provide in compiler explorer, without going through the trouble of a project template. Naturally on a real project there would be an heavily customised static analysis tool, that would only allow a build to succeed with the feedback from the SecDevOps team, alongside feedback loop from pentes…

Instead of ranting and showing a huge warnings output to make a point fitting your agenda, you could have just disabled the false positives yourself (like I did, by the way) and you would have seen that that vastly reduces the warnings.

Oh, and to disprove your other claim, here is a link to the godbolt with added clang-tidy flag: https://godbolt.org/z/G31Ws8aa1 . This has the clang-tidy invocation changed to disable a single warning category : --checks='-clang-analyzer-security.insecureAPI.DeprecatedOrUnsafeBufferHandling' . Running with that, there remains only a single warning. Which is probably a false positive as well.

If there are real concerns about this code, show them. I'm not saying there can't be any. But it doesn't help your credibility if you continue arguing your claims with evidence that is easily disproved. I have nothing against tooling that actually improve the situation. Btw. that `lint` from almost 50 years ago that you're referencing is probably easily covered by `-Wall` or `-Wextra` alone. I was also mentioning valgrind.

Bottom line, you're vastly exaggerating the gravity of the memory bugs inflicted upon us by memory-unsafe languages, compared to other bugs which exist too. (Maybe I like the term memory-dangerous better).

Re: Show HN: A fast, dependency-free traceroute implementation in pure C

#30

Post by pjmlp »

jstimpfle wrote: post #29
pjmlp wrote: post #26

Earlier quoted context omitted.

Instead of ranting, you should have realized that is the default output without configuration file, which isn't that easily to provide in compiler explorer, without going through the trouble of a project template. Naturally on a real project there would be an heavily customised static analysis tool, that would only allow a build to succeed with the feedback from the SecDevOps team, alongside feedback loop from pentes…

Instead of ranting and showing a huge warnings output to make a point fitting your agenda, you could have just disabled the false positives yourself (like I did, by the way) and you would have seen that that vastly reduces the warnings. Oh, and to disprove your other claim, here is a link to the godbolt with added clang-tidy flag: https://godbolt.org/z/G31Ws8aa1 . This has the clang-tidy invocation changed to disable…

Were they false positives though?

How come "which isn't that easily to provide in compiler explorer, " suddenly becomes me telling that was impossible?

If you enjoy typing a endless list of clang-tidy flags on a tiny text box, well fun is on the user.

The first concern is that some parts of the industry keep reaching to C when there are better alternatives, even the language authors have moved on to creating better languages.

Post reply on HN