I had the same problem until I changed how I wrote comments in my code. Instead of using // Some thing, I started using curly braces like this: {/* Some thing */}.

This issue started happening after I upgraded to Expo SDK 52. If you're facing the same error, try updating your comments to this format and see if it helps!

Now with Copilot, you can also ask, and the answer it gives matches the chosen answer on this page (albeit I added quarto so it works in a .qmd file too).

Help...I am in R studio. The torch package is not functioning and the keras and tensorflow not yet.

How can I use ReLu using neuralnet? Is there another form?

If I define relu as cited above, and after I try to use neuralnet and act.fct "relu" not function either. Does anybody have any reference?

The first question can be solved as: =COUNTIF(B$1:B1; B1) (paste to A1 and fill the column)

Count all occurrences of the thing in the neighboring cell starting from row 1 ($1) to the current row.

AUTH_USER_MODEL was commented out in settings.py

If you're using Intellij, then inside the ProjectFolder/.idea/misc.xml

In there you'll see something like languageLevel="JDK_21" followed by more info

Update

From 7.3.1 versions onwards , NTLM is supported in the MS SQL driver

Connection String will look like this

jdbc:sqlserver://<server>;integratedSecurity=true;authenticationScheme=NTLM;domain=<domain>;user=<domainUser>;password=<domainUserPassword>;

It works this way:

import commands as cm

dct = cm.cmd_list

print(dct["one"])

Ok, I managed to resolve the problem. When I changed the image version to

eclipse-mosquitto:latest

it started to work fine. In this image there is mosquitto version 2.0.21, while in

eclipse-mosquitto:2

mosquitto is in version 2.0.18 and it doesn't work.

You can pass cmin = min and cmax = max inside add_surface() to get a consistent color scale across all plots:

add_surface(
    z = z_matrix,
    x = p1,
    y = p2,
    colorscale = "Viridis",
    cmin = 0,
    cmax = 4
)

Had struggled with this today myself, so in case anyone will have the same problem:

You need to set mediaItem property with current active MediaItem.

As option, inside your AudioHandler add this:

in constructor add:

_listenForDurationChanges

void _listenForDurationChanges() {
    _player.durationStream.listen((duration) {
      final index = _player.currentIndex;
      final newQueue = queue.value;
      if (index == null || newQueue.isEmpty) return;
      final oldMediaItem = newQueue[index];
      final newMediaItem = oldMediaItem.copyWith(duration: duration);
      newQueue[index] = newMediaItem;
      queue.add(newQueue);
      mediaItem.add(newMediaItem) //without this notification will be empty; 
    });
  }

Try to update the Node.js version. I ve tried everything and only that worked for me.

MD Arafat Hossain Shiam.

    if ($this->journal3->settings->get('quickCheckoutAccountTelephoneField') === 'required') {
            $telephone = Arr::get($this->request->post, 'order_data.telephone');
    
        if ((utf8_strlen($telephone) == 10 ) || preg_match('/[^\d]/is', $telephone)) {
            $error['telephone'] = $this->language->get('error_telephone');
            }
        }

in opencart 2,3,0,2 Journal theme has another register.php, this works for validation for telephone.

    cmd_value = cmd_list.get(cmd)
#write this

https://github.com/facebook/react-native/issues/49245#issuecomment-2665527685

Check that the following line:

SWIFT_ACTIVE_COMPILATION_CONDITIONS = "$(inherited) DEBUG";

has been added inside project.pbxproj in debug XCBuildConfiguration block:

83CBBA201A601CBA00E9B192 /* Debug / = {
  isa = XCBuildConfiguration;
  buildSettings = {
    ALWAYS_SEARCH_USER_PATHS = NO;
    CLANG_ANALYZER_LOCALIZABILITY_NONLOCALIZED = YES;
    CLANG_CXX_LANGUAGE_STANDARD = "c++20";
    ...
    REACT_NATIVE_PATH = "${PODS_ROOT}/../../node_modules/react-native";
    SDKROOT = iphoneos;
    SWIFT_ACTIVE_COMPILATION_CONDITIONS = "$(inherited) DEBUG";
    USE_HERMES = true;
  };
  name = Debug;
};

Can you also share the compiled javascript, the version of ReScript you are using, and the contents of the rescript.json file?

i have faced the same issue on my mac
Solution that worked for me:

• settings -> display size to more size

• then emulator will be shown bring it to bottom of screen

• now set the display size to default

• now its fixed, drag and drop selected item any way you want

Coming back after searching countless online solutions, this one works the best: https://www.youtube.com/watch?v=Ug7EQairnrA

The author adds withAnimation to manually alter the tab bar visibility that it works without any problem on my side.

Same problem happened to me.

"expo": "~52.0.32", "expo-dev-client": "~5.0.12"

Those versions fixed my problem. Also I have discovered that If I use an QR Scanner application instead of device own camera with SDK-52 it still failing. So install those versions and then scan your QR code with your device's built in camera app.Hope it will help.

You might try the --coverage.include <pattern> CLI argument instead.

https://vitest.dev/guide/cli.html#coverage-include

I encoutered a similar issue. In my case, the reason was that mylocal repo was too large.

So I integrated the old history into a single commit using https://stackoverflow.com/a/50996201/22434761 .

Then, I succeeded in pushing my local repo to CodeCommit.

No, this is way beyond the capability of DTD validation.

My best guess is that you're unable to locate the correct user, possibly because the game property doesn't match the value you're expecting. I'd recommend adding an else statement after the if's where you're trying to find the user and logging the fact that you were unable to locate the user. This should help narrow down where it's going wrong.

After struggling with installation through terminals, I found an easy solution:
https://build-system.fman.io/qt-designer-download

Installed Python 3.13.2 64-bit
Windows 10 64-bit

After you create a snapshot of the disk from the “instance-template” , the missing step is you need to create a boot disk from a snapshot by:

gcloud compute disks create (DISK_NAME) --source-snapshot (SNAPSHOT_NAME) --type=hyperdisk-balanced

Then create a new instance by EXISTING DISKS option then choose the created disk above.

I am facing the same issue.

Please let me know if this issue is resolved?

Thanks,

Anu

We (OpenLink Software) provide a very sophisticated ODBC-JDBC Driver that provides JDBC-accessible data sources with ODBC access from a very broad spectrum of ODBC-compliant applications and services.

These drivers exist in single-tier and multi-tier architecture forms while also being available for on a free evaluation basis -- which should help you make your final decision, cost-effectively.

Upgrading to 2sxc v18 resolved the issue as recommended in the following GitHub issue: https://github.com/2sic/2sxc/issues/3583

As every so often, a trivial mistake: The position parameters for the origin were not converted to int, but were floating point values. I can't believe the old laptop had a resolution which never resulted in .5 floats, but applying int(...) solves the issue.

Switching to wx.GenericStaticBitmap didn't help, by the way.

For me the issue was caused by using:

import { type Route } from "./+types/post"; ❌ wrong

instead of:

import type { Route } from "./+types/post"; ✅ good

Follow the solution of Afshin Rahmati. and add .name like bellow

export class SignUpDto{
    @ApiProperty()
    @MinLength(5, {
        message: i18nValidationMessage('i18n.MIN', {message: 'err'}).name
    })
    username:string;
}

You can use the .getPassword() method to get a password from a JPasswordField!

Although the engineering team has an enhancement request to consider this for a future roadmap, there is currently no way to change the size constraints of a drawn signature. The alternative is to use the stamped name signature and disable the ability to draw signatures.

try this

public int x;
 
public SampleClass(int parameter) : this()
            {
                this.x = parameter;
            }

ThirdMapped is {} that is not empty object but object, anything not null or undefined.

You can do something you can try to solve it with: type EmptyObject = Record<string, never>; type NormalizeEmpty = keyof T extends never ? EmptyObject : T;

gl

In Qt6, if you use cmake rather than configure (both are available), you can pass -DQT_LIBINFIX=_MyInFix to cmake.

=date(if(month(b2)=12,year(b2)+1,year(b2)), if(month(b2)=12,1,month(b2)+1),day(b2))

Using the scripts onload and onerror events is most likely an option here I suppose.

  
  // Set up the onload handler before appending to DOM
  tag.onload = () => {
  //Check if Google Pay exists
  if ('google' in window && window.google.payments && window.google.payments.api) {
      loadGooglePay();
    } else {
      console.error("Where response???");
    }
  };

MongoDB recently released an official integration for Django and MongoDB with out of the box support for key Django features like the ORM, Migrations, Admin Panel etc. It also support all Django data types + most of the MongoDB data types including the Boolean type. I would check it out.

https://github.com/mongodb/django-mongodb-backend

What about a Common Table Expression ?

WITH b AS (
    SELECT b.id, b.column1, b.column2, func(b.id) AS func_result
    FROM Blog b
)
SELECT p.*, b.*
FROM b
JOIN Post p ON b.id = p.b_id;

Ok, the solution was in this page :

The type name 'SerialPort' could not be found in the namespace 'System.IO.Ports'

citing this page :

https://www.nuget.org/packages/System.IO.Ports/

In VS code I had to open the terminal (menu View/>Terminal) and write the line

dotnet add package System.IO.Ports --version 9.0.2

Thanks to the comment from jcalz and the answer from Alexander Nenashev, I've done a "separation of concerns" and come up with what I think is a workable solution here.

You need to compile the project and build package and deploy to Azure Service. You can access that page using local host server.

Running the script as Administrator worked. Interesting that the script (and 6 others like it) ran on the same Windows PC for 8 months without needing to run with Administrator privileges. I tried running the scripts on another PC and they worked fine. So, something (Windows update?) changed that required Administrator privileges. Thanks for all suggestions

The following response headers, that we were manually adding, were causing issues:

x-content-type-options
x-frame-options
x-xss-protection

We removed them and the ERR_BLOCKED_BY_ORB in Chrome went away.

You’ll need an older version of Xcode (Xcode 10 or 11) since the latest Xcode versions don’t support iOS 12.It’s better to use Objective-C or an older Swift version...

Alternatively,

kernprof -lvr script_to_profile.py

to view rich outputs directly

So after some thinking I came to the decision, that efi variables are the better way to go for me.

I just need 9 values and I map them from 0 to 9. I created a efi variable with 9 words separated by spaces and just implemented a function which saves the words to an array.

For a greater application I still would recommend using a config file for the ease of use, but writing a parser for just some values would be overkill.

You can look at my test application here:
https://gist.github.com/KastenKlicker/a1947bec5254d16646dd88484ce8b59e

I don't think this is trivial, but maybe read:

https://arxiv.org/html/2501.10257v1

If you do not have your own api token, you are using the global project that google provides for anonymous testing. This may mean that someone else testing is using the 240 tokens/minute allocated to the test google project.

Simply create a google api token in your developers console for the pageSpeed API. This will also turn the service on.

The below link will get you started

https://developers.google.com/speed/docs/insights/v5/get-started

Try Boxing Barcode:

https://en.wikipedia.org/wiki/Boxing_barcode

https://github.com/piql/boxing

It consists of 4096 rows and 2160 columns. A lot of place for data.

There is a package called corrplot for R and it can handle many variables well.

Its easy to handle and has a lot of options.

Good intro:

https://cran.r-project.org/web/packages/corrplot/vignettes/corrplot-intro.html

For example:

### Loading package 
install.packages("corrplot")
library(corrplot)
### Make some data - only numeric 
data_example <- matrix(rnorm(30*1000), ncol=30)
### Calculate correlation with base R 
M <- cor(data_example)
M
### Create plot with corrplot
corrplot(M, method = 'square', order = 'FPC', type = 'lower', diag = FALSE)

If absolutly nothing is working but you are getting a successful green notification. look to the top right corner of the query page. it looks like a circle/refresh (button) click it. it will reset your layout so the columns and data can show.

I had a very similar problem to what you describe, and finally tracked it down today. In my scenario, I wanted to use API Gateway to deliver requests to a private EC2 instance running LAMP stack and WordPress, and needed access to a custom API at https://wordpress.example.com/api/product which is currently only accessible on our private network.

I finally found that Apache HTTPD is redirecting me to a path with a trailing / so it was trying to redirect me to https://wordpress.example.com/api/product/. I changed my Endpoint URL to have the trailing slash, and that resolved the redirecting issue.

Reference: https://stackoverflow.com/a/64533237/2329657

This can be done by using history api of JS. history.pushState()

this history api provide us a pushState method which can take a data, title and a url where it should redirect. this is used by react-router-dom behind the scene too

This solution blew my mind. Never knew we can loop and union all without creating multiple datasets and this reduced my query time a great deal.

@JasonHarper delivered the simple answer.

encoder = RotaryEncoder(a=14, b=15, max_steps=1000)
encoder.value=.5

value and max_steps are both described as properties in the documentation. max_steps can be an argument. value needs a separate assignment as above. The typography on the man page is unclear. Despite being a property, max_steps cannot be defined outside RotaryEncoder; there is no setter method.

Wow, I'm a little embarrassed that I didn't notice this earlier, but there is a Podfile.properties.json file in the ios directory of the auto-generated project I was completely unaware of until now that was overriding the jsengine as "jsc" (hence the warning indicating hermes engine was not enabled). Simply had to set "expo.jsEngine": "hermes" in that file to enable debugging with Chrome Dev Tools for the ios side of the project.

In Next.js 13.5 and above the syntax has changed. The function properties object now comes second.

Before Next.js 13.5:

  router.push("<new_url>", undefined, { scroll: false })

Next.js 13.5 and above:

router.push("<new_url>", { scroll: false })

Next.js Official docs: https://nextjs.org/docs/app/api-reference/functions/use-router#disabling-scroll-to-top

Credit: https://stackoverflow.com/a/77370718/15903427

var fileSystem = new System.IO.Abstractions.FileSystem();
// Mock System.IO.Abstractions's FileSystem
var mockFS = new Mock<FileSystem>();
var stream = fileSystem.File.Create("my file path.txt");
mockFS.Setup(x => x.File.OpenRead("my file path.txt")).Returns(stream);

Found this answer posted by Super Jade (A million thanks)

How do I create a FileSystemStream?

However, when I test the production code the instance is null. My file does exist at the filePath..

Please set the CSS for your snippet-box-container's align-items property to flex-start.

.snippet-box-container{ display: flex; align-items: flex-start; }                                                                          

Please see this link for more help: https://developer.mozilla.org/en-US/docs/Web/CSS/CSS_flexible_box_layout/Basic_concepts_of_flexbox#align-items

Selecting the first cell with the array formula, along with the target cells to fill this formula to, then pressing Ctrl-D is able to fill the array formula, with the appropriate cell references, without needing to retype it

When using the App Router, we cannot import the router from next/router. Instead, we must use: import { useRouter } from "next/navigation";.

In this case, the router.push function accepts two parameters: router.push(href, options);

So, in my case, I need to update my code like this:

useEffect(() => {
    if (!pageNumber || pageNumber === null || pageNumber === "") {
        router.push(`?pageNumber=1`, { shallow: true, scroll: false });
    }
}, []);

With this change, everything works perfectly, and the page no longer scrolls to the top.

Thanks for the great answer @Vítor Oliveira.

In the lower right you always see the current encoding and can change it as well.
Further the line endings needed to be changed for me from CRLF to LF to match unix systems.

Short addition: Now ANSI is called "Western (ISO 8859-1)" in VS Code.

I just saw this old question and I felt compelled to mention the book Continuous Delivery in Java, which covers precisely this kind of thing.

Full disclosure: I'm one of the co-authors, so obviously biased 😊

I found a workarround using Box and setting the color on LinearProgress to inherit:

import { Box, LinearProgress } from '@mui/material';

export default function ProgressBar() {
  return (
    <Box color="progress.501">
      <LinearProgress
        variant="determinate"
        value={50}
        color="inherit"
        sx={{
          height: 8,
          width: 120,
        }}
      />
    </Box\>
  );
}

However, I wonder if there's a way to make it work like it does in Typography, i.e., without needing an extra Box component.

import cv2

import numpy as np

from PIL import Image

# Load the image

image_path = "/mnt/data/file-1ZEJNRs1fRw6rjpuCRzwQt"

image = cv2.imread(image_path)

# Convert to grayscale

gray_image = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)

# Invert the grayscale image

inverted_image = cv2.bitwise_not(gray_image)

# Apply Gaussian blur

blurred = cv2.GaussianBlur(inverted_image, (21, 21), sigmaX=0, sigmaY=0)

# Invert the blurred image

inverted_blurred = cv2.bitwise_not(blurred)

# Create the pencil sketch

sketch = cv2.divide(gray_image, inverted_blurred, scale=256.0)

# Save the sketch

sketch_path = "/mnt/data/sketch.png"

cv2.imwrite(sketch_path, sketch)

# Return the path

to the sketch

sketch_path

There can be 3 reasons behind it:

1. Try clearing the npm cache:

npm cache clean --force
npm i -g nodemon

2. If your npm is outdated then update it:

npm install -g npm

3. Reinstall your node.js. It might be broken.

if you're using Avast Antivirus, then disable for 10 minutes and run composer again.
It works for me

The SymmetricDS service on Linux needs to be installed first:
bin/sym_service install

See SymmetricDS User Guide for Linux service: https://symmetricds.sourceforge.net/doc/3.15/html/user-guide.html#_running_as_a_linuxunix_daemon

Normally GET is used to list data. I had a similar problem with real-time data working with REACT and PHP. The solution was to change the GET to POST and it worked. I hope this little tip helps you.

I solved this by starting up the installer and repairing my installation.

GCC is a monolithic design architecture, while llvm is more modular.
LLVM has fix the m*n problem of the compiler design to m+n to know you can visit to the blog

https://www.compilersutra.com/docs/llvm/llvm_basic/Why_What_Is_LLVM

While GCC is still used in the legacy software.
To know more about the difference between gcc and llvm, please visit

https://www.compilersutra.com/docs/compilers/gcc_vs_llvm

Sorry about that, the below works of course; too little coffee and was expecting to see the ref in chrome dev mode, jesus.

This works:

const logoEl = useTemplateRef('thelogo')

onMounted(() => {
    console.log(logoEl.value?.$el)

})

</script>

<template>
    <div class="home-wrapper">
        <section class="logo" ref="logo">
            <!--<TheLogo class="logo__svg" :fontControlled="false" />-->
            <NuxtImg src="/thelogo.png" ref="thelogo" loading="lazy" format="webp" quality="100"
                sizes="50vw sm:60vw md:500px" densities="x1 x2" />
        </section>

        <section class="title">THOMAS<br>THORSTENSSON</section>
    </div>
</template>

Thanks, Thomas

OK - so this sort of works but gives me double entries in the box plot legend ...

import holoviews as hv
import hvplot.pandas
import pandas as pd

data = pd.read_csv('https://raw.githubusercontent.com/ChrisWalshaw/DataViz/master/Data/Products/DailySales.csv', index_col=0)
data.index = pd.to_datetime(data.index)
print(data.head())

selected = ['A', 'F', 'L']

plot1 = data[selected].hvplot.line(
    frame_height=300, frame_width=300,
)
plot2 = data[selected].hvplot.hist(
    frame_height=300, frame_width=300,
)
plot3 = data[selected].melt().hvplot.box(
    frame_height=300, frame_width=300, by='variable', c='variable',
    cmap=hv.Cycle.default_cycles['default_colors'],
)
plot = plot1 + plot2 + plot3
hvplot.show(plot)

You can try add a { delay: null } in the userEvent.type(searchComp, 'ABC');:

userEvent.type(searchComp, 'ABC', { delay: null });

Work for me in some cases, I'm using react 18, jest 29 and @testing-library/user-event 14

That would’ve been a great feature, I really don’t know why postman doesn’t have such a feature, it Would have been similar to jetbrain’s ide’s ssh feature for service like logging into a database, for example you only have localhost login to your database, the ide logs with ssh into the machine then you login to the database locally, or if running test docker containers on remote machines you ssh to your remote machines and build and test your docker dev container on that machine

What I ended up doing is building the request in postman and i copy it as a curl code snippet and i run it in my ssh terminal

Or You can follow this article how to use your ssh login as a local http proxy server, postman support http forward proxies

Use OCR and AI-based image processing for tyre text detection and recognition. https://nyumahmobiletyres247.co.uk/

I'm having issues when I use 100vh or h-screen. I haven't found a solution I like, but I noticed that if I changed it to 90vh it would work the way I want better.

Flutter 3.27 fixes it. Hurray!

Search for "Edge to edge" in the announcement blog post and you'll find the news.

Wondering if you ever figured this out? I'm trying to accomplish the same thing. I was able to get the third level to show up but it isn't quite working as expected.

I am doing this at the end of my on_release function to reset the state of the pressed keys:

keyboard._pressed_events.clear()

Thank you for posting and identifying the problem, OP! This was driving me crazy.

This one worked for me (integer numbers only):

([1-9][0-9]+|[1-9])

https://regexr.com/3annd

i am new to amazon connect. I am trying to make a audio works from play prompts or lex messages on audio chat for user that starts webchat. The WEBRTC broswer connection is okay. After calling the active call stays for 5-7 seconds. That is fine as for basic all i want is playpromt audio should get listen . But no audio comes from amazon connect servers . The cloud watch the first log entry of start media stream is fine Butthe 2nd log entry of startmediastreaming shows error in result

"Parameters": {
        "MediaStreamTypes": "Audio",
        "Track": [
            "ToCustomer",
            "FromCustomer"
        ]
    }

I have enable media streaming keneis etc in instance for needful IAM rules ( That managed by AWS anayway) Are you saying listening to audio of play prompts or lex messages are not possible. I have to claim phone number for calls?

I am having the same unresolved reference to io issue in the same file. Did you manage to get it fixed? If yes, please answer your own question. Thanks!

I've found a solution here: https://swiftui-lab.com/matchedgeometryeffect-part1/

Basically what's needed is to use this 'invisible' transition:

extension AnyTransition {
    static var invisible: AnyTransition {
        AnyTransition.modifier(
            active: InvisibleModifier(pct: 0),
            identity: InvisibleModifier(pct: 1)
        )
    }
    
    struct InvisibleModifier: AnimatableModifier {
        var pct: Double
        
        var animatableData: Double {
            get { pct }
            set { pct = newValue }
        }
        
        
        func body(content: Content) -> some View {
            content.opacity(pct == 1.0 ? 1 : 0)
        }
    }
}

i have a same problem. Have you solved the problem?

Warp supports being opened as an external terminal from VS Code with a shortcut. See more in the docs. https://docs.warp.dev/features/integrations-and-plugins#vscode

Warp also has an open feature request for VS Code embedding here: https://github.com/warpdotdev/Warp/issues/257

pip install -q ffmpeg-python

-q flag for installing quietly, as it needs a yes confirmation in proceeding.

When I uninstalled it on my system and then installed the latest version, then the problem disappeared.

I had the same issue, just remove the last update and went back to the last one. Seems to have solved it so far.

I know this is super late, but

in Normal mode:

g?

it will rotate by 13 letters. to un-rotate, press g? again to rotate back.

same problem, manually building with QuarkusRestClientBuilder.newBuilder() does not supports fault tolerance annotations.

Variadic Friends has been implemented in clang 20 now.

https://releases.llvm.org/20.1.0/tools/clang/docs/ReleaseNotes.html

Gcc should support it from gcc 15.
https://gcc.gnu.org/projects/cxx-status.html

P2893R3

I found a solution. I changed the TypeScriptCompileBlocked setting in the project to true. true Thank you to those who responded!

How did you solve it? thanks for your help

I guess some process spawned by the command does change the current directory. But that change does not affect the calling shell's current directory. I did not verify this answer but I hope this helps.

Simple way to fixed it you need edit .env or .env.local file:

NEXT_PUBLIC_SANITY_PROJECT_ID=
NEXT_PUBLIC_SANITY_DATASET=production
NEXT_PUBLIC_SANITY_API_VERSION=
SANITY_STUDIO_PROJECT_ID=
SANITY_STUDIO_DATASET=production

need to duplicate it. with NEXT and without NEXT

• run again sanity deploy;

Good question, Raj!

You can define a "virtual" edge by reusing the existing node table `Interest`. This technique is recommended by us at the user guide: https://cloud.google.com/spanner/docs/graph/best-practices-designing-schema#merging-node-and-edge-input-tables

CREATE PROPERTY GRAPH TestGraph
NODE TABLES (
  InterestCategory,
  Interest,
)
EDGE TABLES (
  Interest AS ContainsInterest
    SOURCE KEY (interest_category_id) REFERENCES InterestCategory
    DESTINATION KEY (interest_id) REFERENCES Interest
)

Query will appear as

GRAPH TestGraph
MATCH p = (n:InterestCategory)-[:ContainsInterest]->(b:Interest)
RETURN TO_JSON(p) as interest_paths;

Given that the two tables have FK reference constraint and shares the prefix key, you can employ INTERLEAVE IN PARENT technique to optimize the schema design, as we recommended here in the user guide: https://cloud.google.com/spanner/docs/graph/best-practices-designing-schema#optimize-forward-edge-traversal