The post is linked here, but to summarise, the task is to convert a string spelled out in NATO phonetic alphabet, without any delimiter, back to a string spelled normally, so for example, from “ECHOXRAYALFAMIKEPAPALIMAECHO” to “EXAMPLE”.

I decided to attempt this in the RISC-V ISA, for no particular reason other than that I have been using it at work. My current score is 52 bytes.

A few details before getting to the code:

• Instead of a full program handling I/O, I will only be implementing a function, using the standard RISC-V ABI.
• I am using the 32-bit ISA with the Zca and Zcb extensions.
• I will assume C strings (i.e. NUL-terminated array of chars) in ASCII.
• I will assume the caller has allocated a sufficiently large output buffer.

My solution relies on the fact that all phonetic alphabets start with the letter it represents, for example, ALFA starts with A, BRAVO starts with B. So all it needs to do is know how long each phonetic alphabet is, and just copy the first letter and skip over the rest.

Here’s the assembly:

    .globl revnato
revnato:
    la a3, revnato_data - 65
revnato_loop:
    lbu a2, 0(a0)
    sb a2, 0(a1)
    beqz a2, revnato_ret
    add a2, a2, a3
    lbu a2, 0(a2)
    add a0, a0, a2
    addi a1, a1, 1
    j revnato_loop
revnato_ret:
    ret
revnato_data:
    .byte 4,5,7,5,4,7,4,5,5,7,4,4,4,8,5,4,6,5,6,5,7,6,7,4,6,4

If you’re not familiar with the syntax, all operations are in the form opcode destination operand or opcode destination operand1 operand2.

Here, a0 is expected to contain the input string, and a1 the output buffer. We first load the address (la) of a lookup table to a3. This table is the string length of each phonetic alphabet. The - 65 will be explained later.

In the loop, we first load one unsigned byte (lbu) at a0[0] into a2, and store that byte (sb) into a1[0], i.e. copy the first character. We exit from the loop (and return from the function) if the byte is NUL (beqz: branch if equal to zero).

If non-zero, we compute the index for the lookup table. The unoptimised way would be to subtract 'A' (ASCII 65) from the byte, and then add it to the lookup table address. To optimise this, I have moved the “subtract 65” to the address itself, so that we can just add a3 to a2. We then dereference the address by loading the unsigned byte.*

We then increment the input pointer by the lookup table entry, and increment the output pointer by 1, and loop.

*: Note that another way to optimise it would be to leave the a3 address as is, and do lbu a2, -65(a2) instead. This may be more readable, but actually produces bigger code, because -65 is outside the range of the 16-bit lbu immediate opcode range (see RISC-V Zcb), so a 32-bit lbu has to be used instead.

This can be compiled to an object file with:

YOUR-TOOLCHAIN-gcc -mabi=ilp32e -march=rv32emzca_zcb -c -o revnato.o revnato.s

We can look at the disassembly to check the binary size:†

YOUR-TOOLCHAIN-objdump -S revnato.o > revnato.dump

00000000 <revnato>:
   0:	00000697          	auipc	a3,0x0
   4:	00068693          	mv	a3,a3

00000008 <revnato_loop>:
   8:	8110                	lbu	a2,0(a0)
   a:	8990                	sb	a2,0(a1)
   c:	c611                	beqz	a2,18 <revnato_ret>
   e:	9636                	add	a2,a2,a3
  10:	8210                	lbu	a2,0(a2)
  12:	9532                	add	a0,a0,a2
  14:	0585                	addi	a1,a1,1
  16:	bfcd                	j	8 <revnato_loop>

00000018 <revnato_ret>:
  18:	8082                	ret

0000001a <revnato_data>:
  1a:	05070504          	.word	0x05070504
  1e:	05040704          	.word	0x05040704
  22:	04040705          	.word	0x04040705
  26:	04050804          	.word	0x04050804
  2a:	05060506          	.word	0x05060506
  2e:	04070607          	.word	0x04070607
  32:	0406                	.short	0x0406

Note that the mv a3,a3, which looks like a NOP, is actually space reserved for a relocation. This is because we are only creating an object file without linking, so the address of revnato_data is not yet known. You can prove this by generating a linked ELF with YOUR-TOOLCHAIN-gcc -mabi=ilp32e -march=rv32emzca_zcb -o revnato.elf revnato.s -nostdlib and observing in that disassembly that the instruction has now changed, for example, to addi a3,a3,-39.

†: Alternatively, we can just run readelf -S revnato.o and look at the size field of .text.

Finally, just for completeness, an example C program to call this function:

#include <stdio.h>
void revnato(const char *, char *);
int main(void)
{
    const char *in = "ECHOXRAYALFAMIKEPAPALIMAECHO";
    char out[100];
    revnato(in, out);
    puts(out);
    return 0;
}

Gave up after spending £16 and still not getting Senjougahara :/

(Signature)
(Signature)

There are a lot other series. To name a few I’ve watched: Oshi no Ko, KiseKoi, Frieren, Spy × Family, TenSura, Cardcaptor Sakura.

This is basically a rewrite of this post by Frederick Zhang. That post is excellent, but there have been some changes since 2018. The following is tested on the latest stable Firefox (138.0.1).

First download this Python script. Use it to decompress search.json.mozlz4 (located in the Firefox profile directory): python3 mozlz4a.py -d search.json.mozlz4 search.json

Optionally, format it with python3 -m json.tool.

Add a new object to the "engines" array. My example for Startpage with custom params:

        {
            "id": "38c37483-6e61-4f86-bfaf-2b99ed7d8464",
            "_name": "Startpage (Unfiltered)",
            "_loadPath": "[profile]/searchplugins/startpage-unfiltered.xml",
            "_iconMapObj": {
                "16": "data:image/x-icon;base64,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"
            },
            "_metaData": {
                "loadPathHash": "<SEE BELOW>",
                "alias": "sp",
                "hideOneOffButton": false,
                "order": 8
            },
            "_urls": [
                {
                    "params": [
                        {
                            "name": "query",
                            "value": "{searchTerms}"
                        },
                        {
                            "name": "cat",
                            "value": "web"
                        },
                        {
                            "name": "lui",
                            "value": "english"
                        },
                        {
                            "name": "prfe",
                            "value": "c774daad5dd23ae2db70252aa321b9354508577d11ee4a79664d62c5d56672dd946c593d0daa5e9f3e7d1ebafaf773669d191e7b802605a743232b31a2254feeccf8ccc0f306bf45f8fa73f3"
                        }
                    ],
                    "rels": [],
                    "template": "https://www.startpage.com/sp/search",
                    "method": "POST"
                }
            ],
            "_orderHint": null,
            "_telemetryId": null,
            "_filePath": "<ABSOLUTE PATH TO PROFILE DIRECTORY>/searchplugins/startpage-unfiltered.xml",
            "_definedAliases": [],
            "_updateInterval": null,
            "_updateURL": null
        }

I assume the "id" is just some UUID. I ended up reusing an existing one that I no longer needed. I’m not sure whether it would be “validated” in any way.

The "loadPathHash" is calculated using the following function:

 function getVerificationHash(name, profileDir = PathUtils.profileDir) {
   let disclaimer =
     "By modifying this file, I agree that I am doing so " +
     "only within $appName itself, using official, user-driven search " +
     "engine selection processes, and in a way which does not circumvent " +
     "user consent. I acknowledge that any attempt to change this file " +
     "from outside of $appName is a malicious act, and will be responded " +
     "to accordingly.";

  let salt =
    PathUtils.filename(profileDir) +
    name +
    disclaimer.replace(/\$appName/g, Services.appinfo.name);

  let data = new TextEncoder().encode(salt);
  let hasher = Cc["@mozilla.org/security/hash;1"].createInstance(
    Ci.nsICryptoHash
  );
  hasher.init(hasher.SHA256);
  hasher.update(data, data.length);

  return hasher.finish(true);
}

This code is taken from the Firefox source code here. It is to be pasted into the Browser Console, then called with getVerificationHash("<VALUE IN _loadPath>", "<ABSOLUTE PATH TO PROFILE DIRECTORY>"). The second argument is required if you are running on a different profile.

The "order" is just whichever unused number that comes next.

Save this, then compress it again with: python3 mozlz4a.py search.json search.json.mozlz4.

Finally, create a backup of the original search.json.mozlz4, and replace it with the new one.

See this post on the Manjaro forum for my fix. To summarize:

• I downgraded everything by following the “Downgrading all upgraded packages” section in this article (parsing /var/log/pacman.log and restoring from /var/cache/pacman/pkg/), and uninstalled alsa-ucm-pinephone. I then re-upgraded and manually reinstalled alsa-ucm-pinephone, and it just magically worked.
• I suspect an easier fix will be to just edit /usr/share/alsa/ucm2/conf.d/simple-card/PinePhone.conf and only keep Syntax 2 and the SectionUseCase."HiFi" and SectionUseCase."Voice Call" sections, and remove everything else.

(Signature)
(Signature)
(Signature)
(Signature)

The symptoms

If you are in the lock screen and have the display turned off (this can due to automatic blank screen, or due to manually pressing the power button), then shortly before the automatic suspend timeout, the notification blue LED will light up.

Previously (before the upgrade), if you ignore the notification, after reaching the timeout, the notification will be cleared, and the PinePhone will be suspended. If you turn on the screen before it suspended, it will very briefly show a notification that the phone will be automatically suspended. However the notification will clear itself almost immediately.

After the upgrade, the notification LED stays on, and the phone does not suspend. When the screen is turned on again, it will always display the notification (which will clear itself very quickly). Most notably, because it does not actually suspend, the battery drains itself within a few hours. I’d go to work in the morning (unplugging from the charger at 08:30), arrive back home at 18:00 and the phone has already turned itself off.

The solution

This has been provided by @alaraajavamma:urheiluaki.org from #pinephone.

gsettings set sm.puri.phosh.notifications wakeup-screen-triggers []

Note that this needs to be run on a terminal on the PinePhone. It does not work via ssh.

For future reference, the original value was ['urgency'].

Other observations

Things we tried and found out before reaching the solution.

• Automatic suspend actually works if the display is turned on and not in the the lock screen.
• sudo systemctl suspend - works.
• systemctl suspend - works on a terminal on the PP. Does not work via ssh.

• systemd-inhibit --list gives the following output:

  WHO            UID  USER    PID  COMM           WHAT                                                     WHY                                          MODE
  ModemManager   0    root    3533 ModemManager   sleep                                                    ModemManager needs to reset devices          delay
  NetworkManager 0    root    3513 NetworkManager sleep                                                    NetworkManager needs to turn off networks    delay
  UPower         0    root    3993 upowerd        sleep                                                    Pause device polling                         delay
  eg25manager    0    root    3311 eg25-manager   sleep                                                    eg25manager needs to prepare modem for sleep delay
  manjaro        1000 manjaro 3826 phosh          handle-power-key                                         Phosh handling power key                     block
  manjaro        1000 manjaro 4045 gsd-media-keys handle-power-key:handle-suspend-key:handle-hibernate-key GNOME handling keypresses                    block
  manjaro        1000 manjaro 4045 gsd-media-keys sleep                                                    GNOME handling keypresses                    delay
  manjaro        1000 manjaro 4046 gsd-power      sleep                                                    GNOME needs to lock the screen               delay
  manjaro        1000 manjaro 3826 phosh          sleep                                                    Phosh handling suspend                       delay
  
  9 inhibitors listed.
  

Failed to call ConnectFilters( pSrc, pMPEG1Splitter ). : [0x80004005] Error: 0x80004005

When running on the command line I get this:

0024:fixme:quartz:mpeg_splitter_sink_query_accept Unsupported subtype {e436eb84-524f-11ce-9f53-0020af0ba770}.

Searched online for these messages and found a few different solutions, but the only one that worked for me was this one by jkfloris:

winetricks l3codecx directshow wmp10

And about a minute in, I noticed some strange high frequency in both audio channels. Here’s a sample:

(Signature)

Not sure what this is, I decided to download it to investigate further:

$ youtube-dl -F 'https://www.youtube.com/watch?v=7fSq0_Ljb_s'

One of the output lines is:

140          m4a        audio only tiny  129k , m4a_dash container, mp4a.40.2@129k (44100Hz), 84.38MiB

So:

$ youtube-dl -f 140 'https://www.youtube.com/watch?v=7fSq0_Ljb_s'
$ ffmpeg -i '【GAMABOOKS_コラボ企画】星空古本市開催！【新人VTuber】-7fSq0_Ljb_s.m4a' -ss 02:00 -t 30 gamabooks_cut.wav
$ octave-cli
octave:1> [y, Fs] = audioread('gamabooks_cut.wav');
octave:2> Y = fft(y);
octave:3> plot(abs(Y));

(Signature)

Here’s our high frequency. Now zoom in to find the FFT index number:

(Signature)

The peak is at index 338543. Convert it to hertz (and note the 1-index):

octave:4> (338543-1) / length(y) * Fs
ans =    1.1285e+04

So now we just need a way to suppress this frequency. Searching “mpv audio filter” online tells us that mpv --af=help will list all audio filters available. Looking through the list, I found:

bandreject       Apply a two-pole Butterworth band-reject filter.

So I just need to find out how to configure this filter. Searching online, I eventually came across some scripts that referenced the FFmpeg manual, specifically, the “equalizer” filter: (emphasis mine)

Apply a two-pole peaking equalisation (EQ) filter. With this filter, the signal-level at and around a selected frequency can be increased or decreased, whilst (unlike bandpass and bandreject filters) that at all other frequencies is unchanged.

Sounds like what I want. So to try it, with reference to the manual (the bandwidth and attenuation factor were determined by trial-and-error):

$ mpv --af=equalizer=f=11285:t=h:width=5:g=-100 gamabooks_cut.wav

Sounds a lot better. Now download the video:

$ youtube-dl -f 136 'https://www.youtube.com/watch?v=7fSq0_Ljb_s'

And play it:

$ mpv 【GAMABOOKS_コラボ企画】星空古本市開催！【新人VTuber】-7fSq0_Ljb_s.mp4 --audio-file=【GAMABOOKS_コラボ企画】星空古本市開催！【新人VTuber】-7fSq0_Ljb_s.m4a --af=equalizer=f=11285:t=h:width=5:g=-100

Done!

Veritasium seems to think that “arbitrariness” is the same as “randomness”. Without true randomness, upon repetition, the guard will gain information on how the renumbering method was chosen, thus decreasing the prisoners’ probability of success.

Uh, what? It looks more like a simply supported beam than a cantilever…

Photo: “Cantilever Stone on Glyder Fach” by George Tod, CC BY-SA 2.0 https://www.geograph.org.uk/photo/611071

(Signature)