doc.undo() works in Desktop version of Word but fails in the Online version

Question

doc.undo() works in Desktop version of Word but fails in the Online version

RahulK 20

Problem Statement : There is a Word document which is being modified by multiple people, not simultaenously though(basically supports co-authoring). This includes all redlines, all comments, content controls, customXMLMappings etc. The requirement is to generate a completely sanitized version of the same document which is devoid of any customXMLMappings, content controls, custom document properties, custom metadata etc.

Current Approach : We are doing the following which is working in Desktop version of Word but is failing in online version.

Loading Metadata: Fetching IDs for Content Controls and Custom XML parts.
Disable "Track Changes": Saving the current changeTrackingMode and turning it off. Doing this ensures that the sanitization steps (like deleting text) aren't recorded as visible "redline" edits.
Unlinking: We remove any XML mapping so the box is no longer "connected" to a data source.
Flattening: We run insertText(cc.text, "Replace") to effectively delete the Content Control "container" but keep the text inside it.
Sync: We invoke ctx.sync(), which sends these commands to the Word application to be executed.
Replacing: We replace the content of each XML part with an empty tag (<empty/>).
Deletion: We delete the entire XML part.
Error Handling: Usage of a try/catch block inside the loop. Based on what we found on the web, some XML parts are "built-in" or protected by Word; if the code tries to delete one of these, it will fail. The catch block allows the script to skip protected parts and keep moving.
Accepting Revisions: Invocation of acceptAllRevisions(), thereby putting all "pending" changes into the final text.
Custom Properties Deletion: We delete all Custom Properties.
Storing the "Clean" State: We call getInternalDocumentBase64() to save the currently sanitized state into a variable (which is a Base64 string).
Revert Everything: We loop through a doc.undo() command for every sync() that occurred.
Restoration: We restore the original "Track Changes" setting.
Variable Assignment: The document looks exactly like it did before the function started, but the variable cleanBase64 now holds the sanitized version.
Verification: We create a temporary document in memory using the cleanBase64 data and opens it to fetch the clean sanitized version of the document.

Now, step 12 is where the code starts to fail for the online version of Word. We need to undo the changes, but that throws an exception.

Could you please suggest a resolution for this or suggest an alternative that works seamlessly in both desktop as well as online version of Word.

Viorel 126.7K Reputation points

2026-01-08T17:42:11.59+00:00

(Removed)

2 answers

Your answer

Viorel 126.7K Reputation points

2026-01-08T17:42:11.59+00:00

(Removed)

Answer 1

Austin-H 8,330 Microsoft External Staff Moderator

Hello RahulK
Thank you for posting question in Microsoft Q&A Forum.
According to this issue, I have found this URL: https://learn.microsoft.com/en-us/javascript/api/requirement-sets/word/word-api-desktop-1.4-requirement-set?view=word-js-preview
The WordApiDesktop requirement set is a special requirement set that includes features that are only available for Word on Windows and on Mac. APIs in this requirement set are considered to be production APIs for the Word application on Windows and on Mac. WordApiDesktop APIs are considered to be "preview" APIs for other platforms (for example, web) and may not be supported by any of those platforms.
I also found undo() is displayed on this:
User's image

You could check on this by searching in the table below the URL I provided above.

Based on my understanding, in Word on the web, every user is a part of real time co-authoring session. When you call acceptAllRevisions(), Word on the Web must reconcile this with the "Master" version on the server. To prevent the document from becoming corrupted or out of sync for other users, the web client often clears the local Undo stack. Once the stack is cleared, the undo() command fails because there is no "previous state" to return to.
Instead of using undo() as it is unreliable in the browser, you could consider performing a sanitization on a "copy" of the document data. Instead of changing user's live document, you could take a copy then clean it and save the result.
I also read your step 15 and thought that it should have been the first step of this sanitization process.
After all, my suggestion should be using 15th step as the first step as the undo() has its own limitation.
I hope this information is helpful. Please let me know if I misunderstood your point at any time or if you have any further question.

If the answer is helpful, please click "Accept Answer" and kindly upvote it. If you have extra questions about this answer, please click "Comment".

Note: Please follow the steps in our documentation to enable e-mail notifications if you want to receive the related email notification for this thread.

Austin-H 8,330 Reputation points Microsoft External Staff Moderator

2026-01-12T00:16:05.22+00:00

Hello RahulK
It has been a while since your last approach.
May I know if you have checked on my answer?
Please let me know if you have any further questions.
RahulK 20 Reputation points

2026-01-12T10:26:22.7333333+00:00

HI Austin,

One thing which i noticed with my existing code is that the flow worked fine for Mac version of Word but it failed for Windows version of word. Seems like Windows word version is more restrictive on what can be used.

Also, i did check with the suggestion you provided to create a document clone first but the issue happening with that is --- the original document is being considered as the current active context document even though we have segregated the cloned document into it's own context.
RahulK 20 Reputation points

2026-01-12T10:34:01.7633333+00:00
Capture the Base64 of the current (dirty) document first.

Open that Base64 in a temporary document context.

Perform all the destructive cleaning (Deleting XML, Flattening CCs) on that temporary document.

Save the final version from the temporary document.

This is the Clone and Modify approach which I tried
Austin-H 8,330 Reputation points Microsoft External Staff Moderator

2026-01-13T06:35:27.5133333+00:00

Hello RahulK
Thank you for your response.
From this part:
"Also, i did check with the suggestion you provided to create a document clone first but the issue happening with that is --- the original document is being considered as the current active context document even though we have segregated the cloned document into it's own context"
Are you able to provide a part of the code? Especially when at the part you selected the document. I would need to see it first to have a research/check on it.
I also initiated a private message if you would like to share it in private with me.
Austin-H 8,330 Reputation points Microsoft External Staff Moderator

2026-01-14T08:06:29.38+00:00
Hello RahulK
Thank you for your response on private message. As the structure within the private message is not well-structured. Please check on my response below.
First of all: check on this article: https://learn.microsoft.com/en-us/office/dev/add-ins/word/create-better-add-ins-for-word-with-office-open-xml
1.Switch to OOXML for clone input
Snapshot the source: const ooxml = ctx.document.body.getOoxml(); await ctx.sync();

Create the clone from OOXML: const tempDoc = ctx.application.createDocument(ooxml.value);
2.Try to operate Word.run(tempDoc, ...) after open()

tempDoc.open(); await ctx.sync(); await Word.run(tempDoc, async (cctx) => { const ccs = cctx.document.contentControls; ccs.load("items/id,text"); await cctx.sync(); // Your code to sanitify the file await cctx.document.save(); await cctx.sync(); });

This should result in edits applying only to the clone.
Please try this and let me know if it could make this different.
Austin-H 8,330 Reputation points Microsoft External Staff Moderator

2026-01-15T07:57:21.48+00:00

Hello RahulK
I hope this comment finds you well.

May I know if you have any updates from your end?
I am looking forward to your response.
RahulK 20 Reputation points

2026-01-16T05:45:09.4233333+00:00

I am currently trying to implement this approach. Will update here with the results, soon.
Austin-H 8,330 Reputation points Microsoft External Staff Moderator

2026-01-16T07:05:16.3966667+00:00

Hello RahulK
Thanks for your response.

I am looking forward to it!
RahulK 20 Reputation points

2026-01-16T08:20:12.84+00:00
I tried the approach you suggested.. Each process adopted is detailed below..

calling tempDoc.open() initially and then doing edits was not working as calling open() was severing the current context link(since it is a terminal action) and I was hitting a GeneralException error.

I switched the open() call invocation to the end. Here's what i did : a. Capture OOXML: Get the data from the original (which remains untouched). b. Create Hidden Doc: Create tempDoc using the OOXML. This exists in the background. c. Modify in Background: Ran the sanitization logic on tempDoc. d. Open at the End: Calling tempDoc.open() only after all modifications are synced.

But I ran into a different problem --- body.getOoxml() was not returning root relationships file(_rels folder post unzipping the docx file) and this was causing createDocument to fail since that requires to treat the XML as a standalone file. I was getting InvalidArgument error in the console.

I managed to bypass the above error using a blank document and then injecting the Ooxml in it. a. // 1. Capture original OOXML const originalOOXML = ctx.document.body.getRange().getOoxml(); b. // 2. Create blank doc and inject content const tempDoc = ctx.application.createDocument(); tempDoc.body.insertOoxml(originalOOXML.value, Word.InsertLocation.replace); c. Loading data for sanitization and clean up d. tempDoc.open();
await ctx.sync();

This worked fine on Mac Word version, I need to test the same for Windows Word version.
Austin-H 8,330 Reputation points Microsoft External Staff Moderator

2026-01-19T06:13:58.17+00:00

Hi RahulK
Thank you for your response.
May I know if you meant that my suggestion combines with your code development results in being able to make the script run on seprate file on Mac but not Windows?
I just want to make sure to make further research on your previous response.
RahulK 20 Reputation points

2026-01-19T10:44:56.92+00:00

HI Austin,

I have detailed the steps that I tried with the results for each posted. As mentioned earlier, approach 3 worked fine for Mac Word.. Need to run on Windows word and check again.
Austin-H 8,330 Reputation points Microsoft External Staff Moderator

2026-01-20T02:47:49.6533333+00:00

Hi RahulK
Thank you for your response.
Once you check this on Word for Windows and have any further update, please let me know, I will proceed to update it in initial response to help with further similar questions.
RahulK 20 Reputation points

2026-01-20T11:02:15.26+00:00

HI Austin,

The Word version of Windows is behaving in the same way as Mac.. However, there is one issue that I am facing. acceptAllRevisions() method does not seem to be working.

I don't see "LOG: [Step 1] Tracked changes accepted" nor do I see "Step 1 Error" in the console logs. Also, in the final document that opens up, it shows changes which have not been accepted. Can you point out what is wrong here?

Have shared the code in private chat.
RahulK 20 Reputation points

2026-01-21T10:47:14.9866667+00:00

Hi Austin,

I managed to fix the acceptAllRevisions() issue. Now the newly created document is a sanitized version of the original document with all changes/revisions accepted. However, there is an issue which I am facing currently, which is the image present in the Header section of the original document is not being copied over to the newly created document. From what I saw, header and its images are not part of the main document body. How do I tackle this as part of the current implementation?
Austin-H 8,330 Reputation points Microsoft External Staff Moderator

2026-01-22T06:53:51.68+00:00
Hi RahulK
Thank you for your response.

I also checked on this GitHub link: https://github.com/OfficeDev/office-js/issues/2623
The idea should be:

Get header OOXML from the original document

Create your blank doc and inject the body OOXML as you already do

Insert header OOXML into the new doc’s header body (repeat for footer if needed)

Do this for each header type (same as footer).
Or you could think of the idea to clone from the full .docx package (the base64 you already know how to build via getFileAsync(Compressed)). The createDocument(base64File) API is explicitly defined to create a new document from a Base64‑encoded .docx file.

Please check the options above to see if anything of that works with you.
Austin-H 8,330 Reputation points Microsoft External Staff Moderator

2026-01-23T07:12:02.3266667+00:00

Hi RahulK
Any updates from your end?
I am looking forward to your response.
RahulK 20 Reputation points

2026-01-27T07:38:42.8066667+00:00

Hi Austin,

Apologies for the delayed response.

As mentioned in the above Github link shared by you specifically this --- https://github.com/OfficeDev/office-js/issues/2623#issuecomment-1256901928, the insertion into the Header is being done using an already extracted XML from a document. In our case, the document will have certain info pertaining to each individual document, and therefore we cannot use one generic header xml string for all documents. And this has to happen at runtime since it is a user triggered action. Please correct me if I am wrong.

Secondly, if you revisit some of the comments at the beginning, we did try to use Base64 but ran into some issue or the other, post which we went ahead with OOXML as per your suggestion.
Austin-H 8,330 Reputation points Microsoft External Staff Moderator

2026-01-28T06:25:42.18+00:00

Hi RahulK
Thank you for your response.
I agree that a single “generic” header XML won’t work when each document has its own header content, but the approach we’re suggesting doesn’t rely on a shared template at all: instead, the add-in extracts the header/footer OOXML from the same document the user is working on at runtime, then inserts that extracted header/footer OOXML into the corresponding header/footer of the newly created sanitized document (in the same way you already insert the body OOXML).

This is consistent with the guidance that header/footer content must be handled separately from the main body and inserted via the header/footer APIs, using a separately obtained OOXML payload rather than document.body.insertOoxml() alone.
Since I don’t have the same exact client build and document variants as you, could you try this “extract and insert header/footer from the current doc” path in your environment and share any error debugInfo if it still fails?
RahulK 20 Reputation points

2026-01-30T06:54:15.04+00:00

Hi Austin,

I managed to make the Header along with the images show up correctly as well based on your above suggestion. I am yet to make the changes for the Footer, but I assume it would be the same and should work as expected.

However, I see one major issue which requires to be fixed - numbering system seems to be completely gone with this approach, they are not retained. The original document has paragraphs which are numbered and paragraph headers act as content controls. From what I read, using cc.getRange().insertText(cc.text, "Replace") is causing this, but I am using this to flatten the Content Controls. Same with fonts, they seem to be all jumbled up.

Any workaround for this?
Austin-H 8,330 Reputation points Microsoft External Staff Moderator

2026-02-02T02:54:02.39+00:00

Hi RahulK
Thank you for your response.

Instead of replacing with insertText, could you try to delete the content control but keep its content using delete(true)?
Based on this article about Content Control: https://learn.microsoft.com/en-us/javascript/api/word/word.contentcontrol?view=word-js-preview#word-word-contentcontrol-delete-member(1)
Please try to approach like this:

Let me know once you have the update from your end.
Deleted

This comment has been deleted due to a violation of our Code of Conduct. The comment was manually reported or identified through automated detection before action was taken. Please refer to our Code of Conduct for more information.
Deleted

This comment has been deleted due to a violation of our Code of Conduct. The comment was manually reported or identified through automated detection before action was taken. Please refer to our Code of Conduct for more information.
Deleted

This comment has been deleted due to a violation of our Code of Conduct. The comment was manually reported or identified through automated detection before action was taken. Please refer to our Code of Conduct for more information.
Deleted

This comment has been deleted due to a violation of our Code of Conduct. The comment was manually reported or identified through automated detection before action was taken. Please refer to our Code of Conduct for more information.
RahulK 20 Reputation points

2026-02-02T08:07:01.0066667+00:00

I did try the delete(true) approach which you suggested above. But that still does not bring back the numbering. Also, it failed to remove a few content controls which are mapped to customXmlParts. Using insertText(), all content controls were getting removed, even the ones mapped to customXmlParts.Screenshot 2026-02-02 at 1.36.23 PM.png
Austin-H 8,330 Reputation points Microsoft External Staff Moderator

2026-02-03T07:33:18.37+00:00

Hi RahulK
Thank you for your response.
How about cloning from the full DOCX package (Base64)? It seems to me that OOXML injection method does not bring the number to new file.

My apology for this inconveniences caused by this issue, but at the moment I could not provide further steps on the OOXML injection method.
Please try to clone from full DOCX package to see how it goes.
RahulK 20 Reputation points

2026-02-03T16:07:37.9366667+00:00

Hi Austin,

I had tried the base64 approach earlier but that was considering the original document as the active document and changes were being reflected in that. You can see the steps I undertook for it in comments 2 and 3, right at the very beginning of this comment section --- Clone and Modify approach. To overcome that, I used OOXML as per your suggestion.
RahulK 20 Reputation points

2026-02-03T16:59:07.4033333+00:00
Hi Austin,

Latest update --- I basically replaced the OOXML with the base64 content and could see two different results.1. If delete(true) is being used, I can see that the content controls have not been removed. They are still there, as evident from the presence of w:sdtContent tags in document.xml.. Code for this is similar to the screenshot shared earlier.

If I use insertText() to remove the content controls, entire document body shows up as redline(header and footer excluded), which is not acceptable.

Please suggest how to proceed next as both OOXML as well as base64 approach have their own challenges and limitations.
RahulK 20 Reputation points

2026-02-05T03:44:32.1833333+00:00

Hi Austin,

Can you please suggest a workaround on this? As mentioned, both OOXML as well as base64 are not working completely. We need the styling/formatting/numbering to work along with the removal of content controls.
Austin-H 8,330 Reputation points Microsoft External Staff Moderator

2026-02-05T04:19:56.26+00:00
Hi RahulK
Thank you for your response and my apology for any delay in answering your question.

Until this moment when I observed your update on this, I have been in thoughts on how to sort this out as each method (Base64 or OOXML) is not giving any expected result from your end.
My last suggestion is to consider removing remaining SDTs that Word API doesn’t surface by doing DOCX package surgery:

Unzip the DOCX (Base64) in code (JSZip or server-side Open XML SDK).

In /word/document.xml (and /word/header*.xml, /word/footer*.xml), replace each <w:sdt> wrapper with its <w:sdtContent> children (i.e., “unwrap” SDTs rather than rewriting text).

Keep /word/numbering.xml, /word/styles.xml, and relationships intact — that’s what preserves numbering and formatting.

Re-zip → Base64 → open as the sanitized copy.

From my perspetive thought on this situation, the conclusion would be: API-only won’t meet 100% of your stated constraints; package-level sanitation will.
I hope this should work, but it's already far from using Office.js as it could not work under your situation.
RahulK 20 Reputation points

2026-02-06T06:00:50.6833333+00:00

Hi Austin,

New update :

The code is working fine in Windows desktop Word but intermittently. The removal of content controls, acceptance of all revisions, formatting, styling, numbering etc all is intact. But as I mentioned, intermittently.

However, on Mac desktop Word, formatting, styling, numbering is distorted.

Could you please share your inputs on this? Please note that usage of JSZip is not an option that we can undertake.
Austin-H 8,330 Reputation points Microsoft External Staff Moderator

2026-02-10T03:30:18.0833333+00:00

Hi RahulK
Thanks for your response.

Did you try to test this with different version of Office desktop for Mac? Are there any cases that this could work on Mac desktop (assuming that you could check this on multiple Mac devices with different versions).
https://github.com/OfficeDev/office-js/issues/6232
I am looking into this issue raised for this and seeing the inconsistency for this for Word for Mac.
I am still thinking of this - but unfortunately nothing has come up for now. Could you raise this issue about this intermittence on Github?
RahulK 20 Reputation points

2026-02-11T09:26:24.9133333+00:00

Hi Austin,

Sure, let me try it out on different versions of Office for Mac and update here.

Another issue which I had is that the code works fine for desktop versions(if you ignore the styling and formatting) but in Word online, right after the header content insertion using insertOOXML ,when we invoke await ctx.sync(), we are getting a RichApiError. Is the above method not supported in Word online?
Austin-H 8,330 Reputation points Microsoft External Staff Moderator

2026-02-12T07:42:20.12+00:00

Hi RahulK
Thanks for your response.
I found this thread as well: https://learn.microsoft.com/en-us/answers/questions/1194433/insertooxml%28%29-method-of-content-control-is-not-wor - this could be similar to your description.

Do you see the actual error looking like this? (Reference: https://github.com/OfficeDev/office-js/issues/4640)
or this: https://github.com/OfficeDev/office-js/issues/4698
It could be possible that this is having issue with the header and still being addressed for a long time.
As I mentioned earlier, it's recommended to push this into GitHub again for further assistance.

Answer 2

The issue you're encountering with the doc.undo() method failing in the online version of Word may be due to the partial support for undo functionality in the Office JavaScript APIs for online applications. Unlike the desktop version, where the undo stack can be manipulated more freely, the online version has limitations that can lead to exceptions when trying to undo certain operations, especially after a series of complex changes like the ones you are performing.

To resolve this, consider the following alternative approaches:

Avoiding Undo: Instead of relying on doc.undo() to revert changes, you can maintain a copy of the document's state before making modifications. This way, you can simply restore the document to its original state by reloading the saved version instead of using undo.
Batch Operations: Try to minimize the number of sync() calls. Group multiple changes together and call ctx.sync() less frequently. This can help reduce the complexity of the undo stack and may prevent exceptions.
Error Handling: Enhance your error handling to manage specific exceptions that may arise from the online version. This could involve checking the type of error thrown and deciding whether to skip the undo operation or take an alternative action.
Simplifying the Process: If possible, simplify the sanitization process by reducing the number of steps or operations that require undo. For example, if certain operations can be combined or eliminated, it may help in avoiding issues with the undo functionality.
Consult Documentation: Keep an eye on the latest updates in the Office JavaScript API documentation, as Microsoft may introduce improvements or changes that could affect how undo operations are handled in the online version.

By implementing these strategies, you should be able to create a more robust solution that works seamlessly across both desktop and online versions of Word.

Share via

doc.undo() works in Desktop version of Word but fails in the Online version

2 answers

Your answer