Invoke-RestMethod parameter combination of OutFile/PassThru does not write to file as intended

[sdwheeler]

Problem description

In Invoke-RestMethod, the PassThru parameter must be used with the OutFile parameter. The intent is that you can write the results to the file and get the output to the pipeline. But the cmdlets only write to the pipeline, not the file.

The problem code starts here:

PowerShell/src/Microsoft.PowerShell.Commands.Utility/commands/utility/WebCmdlet/Common/InvokeRestMethodCommand.Common.cs

Line 387 in 38d582f

if (ShouldWriteToPipeline)

if (ShouldWriteToPipeline)
{
   ...
}
else if (ShouldSaveToOutFile)
{
   ...
}

Notice that the logic outputs pipeline and skips the output to the file. Both ShouldWriteToPipeline and ShouldSaveToOutFile can be true so need to remove the else.

Steps to reproduce

Invoke-RestMethod https://taxonomyservice.azurefd.net/taxonomies/product-uri -PassThru -OutFile C:\temp\irm.json

Expected behavior

Output to console and to the file.

Actual behavior

Output to console only. The output file is not created.

Environment data

This affects all versions 5.1 and higher.

[sdwheeler]

After further investigation, removing the else does not fix the problem. When the else is removed, the outfile is created but empty. The problem is that the stream is consumed by writing to the pipeline. There is no way to rewind the stream so that it can be written to the file as well.

The only "fix" would be to disallow the parameter combination. That would be a breaking change.

[iSazonov]

Why can't we implement this if we can have Tee-Object?

[sdwheeler]

@iSazonov I suppose we could, but I think that would be a large change to the architecture of the cmdlet. @markekraus may have a better answer.

[vexx32]

My inclination would be to write the file and then read the data, perhaps 🤔

Or, alternatively, can we reset the stream's position after writing to the pipeline and then use it to write the file?

[sdwheeler]

@vexx32 Swapping the order does not change the problem. The problem is that the StreamHelper class does not provide a way to rewind the stream. So the helper class needs to change, which is potentially a lot riskier.

EDIT: Oh, I just realized you are suggesting we write the file first then read the file to output to the pipeline. I still prefer having a way to rewind the stream pointer.

[vexx32]

Yeah. For small files, we could cache the file in memory as a memory stream before saving/outputting, since those allow us to rewind the stream.

To account for large files, it might make sense to use a MemoryStream (or similar) as a buffer (make it say 1024 bytes or whatever) and:

• copy X bytes from the StreamHelper into the MemoryStream
• write all those to the pipeline from the stream
• rewind the MemoryStream
• write the bytes to the file
• repeat the process until there's no more data from the StreamHelper

[iSazonov]

@iSazonov I suppose we could, but I think that would be a large change to the architecture of the cmdlet. @markekraus may have a better answer.

I don't see any complicity to add this - there is only two places there we output.

And we read all in buffer :-) - BufferingStreamReader - and rewind it with responseStream.Seek(0, SeekOrigin.Begin);

[vexx32]

If we have a large file download -- say, 10gb -- does that still work? Or will the buffering mean that one or both modes of output will only get partial data? Can we afford to completely rewind a 10gb stream or similar and force it to be all kept in memory?

Seems to me that we should avoid having the entire stream in memory as much as possible.

[iSazonov]

If we have a large file download -- say, 10gb -- does that still work?

10 Gb for REST? This is not a realistic scenario as far as I'm concerned. Invoke-WebRequest is for downloading large files.

[sdwheeler]

I don't think it will work to try to mix handling of the output together. When you write to the file, it writes the raw data stream. When you write to the pipeline it does a lot of conversion from XML or JSON and creates a PSCustomObject. So pipeline case needs the complete stream before it can transform the data.

The real fix to so be able to rewind the stream pointer. When I talked to @markekraus, he didn't seem to think that was possible with the current StreamHelper class.