{"id":105922,"date":"2021-11-15T07:00:00","date_gmt":"2021-11-15T15:00:00","guid":{"rendered":"https:\/\/devblogs.microsoft.com\/oldnewthing\/?p=105922"},"modified":"2021-11-15T05:16:34","modified_gmt":"2021-11-15T13:16:34","slug":"20211115-00","status":"publish","type":"post","link":"https:\/\/devblogs.microsoft.com\/oldnewthing\/20211115-00\/?p=105922","title":{"rendered":"A practical use for GetHGlobal­FromStream<\/CODE> when sharing was never your intention"},"content":{"rendered":"

A little while ago, I noted that managing shared access to the HGLOBAL<\/code> inside a stream can get tricky<\/a>, and opined that the GetHGlobal\u00adFromStream<\/code> should have been something like IStream\u00adOn\u00adHGlobal::Detach\u00adHGlobal<\/code><\/a>. But in fact there’s a straightforward use case for GetHGlobal\u00adFromStream<\/code> even when there is no sharing going on.<\/p>\n

The STGMEDIUM<\/code> structure is a currency for passing data into and out of an IData\u00adObject<\/code>. it is basically a discriminated union:<\/p>\n

struct STGMEDIUM\r\n{\r\n    DWORD tymed;\r\n    union\r\n    {\r\n        HBITMAP hBitmap;\r\n        HMETAFILEPICT hMetaFilePict;\r\n        HENHMETAFILE hEnhMetaFile;\r\n        HGLOBAL hGlobal;\r\n        LPOLESTR lpszFileName;\r\n        IStream* pstm;\r\n        IStorage* pstg;\r\n    };\r\n    IUnknown* pUnkForRelease;\r\n};\r\n<\/pre>\n
The tymed<\/code> member specifies which of the members of the union is active. Let’s focus on the case where the value is TYMED_MWBR>HGLOBAL<\/code>, in which case the structure simplifies to<\/p>\n
struct STGMEDIUM\r\n{\r\n    DWORD tymed = TYMED_HGLOBAL;\r\n    HGLOBAL hGlobal;\r\n    IUnknown* pUnkForRelease;\r\n};\r\n<\/pre>\n
When you are finished with the STGMEDIUM<\/code>, you call Release\u00adStg\u00adMedium<\/code>, and in the case of a TYMED_HGLOBAL<\/code> the rule for Release\u00adStg\u00adMedium<\/code> is<\/p>\n
    \n
  • If pUnkForRelease<\/code> is not null, then call pUnkForRelease->Release();<\/code>.<\/li>\n
  • If pUnkForRelease<\/code> is null, then call GlobalFree(hGlobal);<\/code>.<\/li>\n<\/ul>\n
    In words, if there is a pUnkForRelease<\/code>, then it is the “owner” which controls the lifetime of the HGLOBAL<\/code>, and you tell it that you are done by releasing the reference count on the owner. When the owner’s reference count goes to zero, it destroys itself (and the HGLOBAL<\/code>).<\/p>\n
    On the other hand, if there is no pUnkForRelease<\/code>, then the HGLOBAL<\/code> is “ownerless”, and you just free it when you’re done.<\/p>\n
    The idea here is that if the HGLOBAL<\/code> was expensive to produce, your data object may decide to cache the result rather than having to produce it from scratch every time. In that case, when you give out the data, you can do this:<\/p>\n
    if (m_cachedHglobal == nullptr)\r\n{\r\n    RETURN_IF_FAILED(Calculate(&m_cachedHglobal));\r\n}\r\n\r\npstgm->tymed = TYMED_HGLOBAL;\r\npstgm->hGlobal = m_cachedHglobal;\r\npstgm->pUnkForRelease = this;\r\nthis->AddRef(); \/\/ because we put it in pUnkForRelease\r\n<\/pre>\n
    When the recipient of the data is finished, they will call Release\u00adStg\u00adMedium<\/code>, and Release\u00adStg\u00adMedium<\/code> will see that there is a non-null pUnkForRelease<\/code> and instead of freeing the HGLOBAL<\/code>, it’ll release the pUnkForRelease<\/code>.<\/p>\n
    This means that the m_cachedHglobal<\/code> is not destroyed when the recipient of the data is finished. It lives on, so it can be returned to another client.<\/p>\n
    Finally, when the data object is destroyed, it also destroys the m_cachedHglobal<\/code>.<\/p>\n
    This pattern means that as long as anybody still has a STGMEDIUM<\/code> referring to your cached HGLOBAL<\/code>, your entire data object will remain alive. But maybe your data object has a lot of stuff in it, like an entire HTML DOM, and the HGLOBAL<\/code> is a cache of the textContent<\/code>. Somebody who asks for the textContent<\/code> and hangs onto it for a long time will keep your data object alive, even if they aren’t using the data object any more:<\/p>\n
    HRESULT GetTheText(STGMEDIUM* pstgm)\r\n{\r\n    wil::com_ptr<IDataObject> pdto;\r\n    RETURN_IF_FAILED(GetTheDataObject(&pdto));\r\n\r\n    FORMATETC fe;\r\n    fe.cfFormat = CF_UNICODETEXT;\r\n    fe.ptd = nullptr;\r\n    fe.dwAspect = DVASPECT_CONTENT;\r\n    fe.lindex = -1;\r\n    fe.tymed = TYMED_HGLOBAL;\r\n    \r\n    RETURN_IF_FAILED(pdto->GetData(&&fe, pstgm));\r\n\r\n    return S_OK;\r\n}\r\n<\/pre>\n
    This function gets the data object and extracts the Unicode text. The data object is thrown away when the com_ptr<\/code> destructs, and all that remains is the text in the STGMEDIUM<\/code>‘s HGLOBAL<\/code>.<\/p>\n
    The catch here is that the caller of this function might decide to keep the text for a long time, and that’s going to keep your big data object around for a long time. Even though all that really needs to be kept alive is the text.<\/p>\n
    This is where you can use one of the lesser powers of GetHGlobal\u00adFromStream<\/code>.<\/p>\n
    Instead of making the data object be the cache for the HGLOBAL<\/code>, you can make an HGLOBAL<\/code>-backed stream with Create\u00adStream\u00adOn\u00adHGlobal<\/code>, and let the stream be the one in charge of the HGLOBAL<\/code>‘s lifetime.<\/p>\n
    if (m_cachedStream.get() == nullptr)\r\n{\r\n    wil::unique_hglobal text;\r\n    RETURN_IF_FAILED(Calculate(&text));\r\n    RETURN_IF_FAILED(CreateStreamOnHGlobal(\r\n        text.get(), TRUE \/* fDeleteOnRelease *\/,\r\n        &m_cachedStream));\r\n    text.release(); \/\/ m_cachedStream owns it now\r\n}\r\n\r\npstgm->tymed = TYMED_HGLOBAL;\r\nRETURN_IF_FAILED(GetHGlobalFromStream(\r\n        m_cachedStream.get(), &pstgm->hGlobal));<\/span>\r\n\/\/ The stream is the owner of the HGLOBAL\r\npstgm->pUnkForRelease = m_cachedStream.copy().detach();<\/span>\r\n<\/pre>\n
    This time, the owner of the HGLOBAL<\/code> is the m_cached\u00adStream<\/code>, and therefore if the storage medium is retained beyond the life of the data object, the data object can destruct, and the m_cachedStream<\/code> will deal with freeing the HGLOBAL<\/code> on final release.<\/p>\n
    I’m guessing this might even have been the scenario for which GetHGlobal\u00adFromStream<\/code> was originally invented.<\/p>\n
    Mind you, we’re using an entire stream just to babysit an HGLOBAL<\/code>. We could have written a custom babysitter:<\/p>\n
    struct IUnknownOnHGLOBAL : winrt::implements<IUnknownOnHGLOBAL, ::IUnknown>\r\n{\r\n    IUnknownOnHGLOBAL(HGLOBAL glob) : m_glob(glob) {}\r\n    wil::unique_hglobal glob;\r\n};\r\n<\/pre>\n
    On the other hand, using Create\u00adStream\u00adOn\u00adHGlobal<\/code> may end up being the easier route if the HGLOBAL<\/code> was originally generated from a stream in the first place:<\/p>\n
    if (m_cachedStream.get() == nullptr)\r\n{\r\n    wil::com_ptr<IStream> stm;\r\n    RETURN_IF_FAILED(CreateStreamOnHGlobal(\r\n        nullptr, TRUE \/* fDeleteOnRelease *\/,\r\n        &stm));\r\n    RETURN_IF_FAILED(SaveToStream(stm.get()));\r\n    m_cachedStream = std::move(stm);<\/span>\r\n}\r\n\r\npstgm->tymed = TYMED_HGLOBAL;\r\nRETURN_IF_FAILED(GetHGlobalFromStream(\r\n        m_cachedStream.get(), &pstgm->hGlobal));\r\npstgm->pUnkForRelease = m_cachedStream.copy().detach();\r\n<\/pre>\n","protected":false},"excerpt":{"rendered":"
    This may even have been its original purpose.<\/p>\n","protected":false},"author":1069,"featured_media":111744,"comment_status":"open","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"_acf_changed":false,"footnotes":""},"categories":[1],"tags":[25],"class_list":["post-105922","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-oldnewthing","tag-code"],"acf":[],"blog_post_summary":"
    This may even have been its original purpose.<\/p>\n","_links":{"self":[{"href":"https:\/\/devblogs.microsoft.com\/oldnewthing\/wp-json\/wp\/v2\/posts\/105922","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/devblogs.microsoft.com\/oldnewthing\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/devblogs.microsoft.com\/oldnewthing\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/devblogs.microsoft.com\/oldnewthing\/wp-json\/wp\/v2\/users\/1069"}],"replies":[{"embeddable":true,"href":"https:\/\/devblogs.microsoft.com\/oldnewthing\/wp-json\/wp\/v2\/comments?post=105922"}],"version-history":[{"count":0,"href":"https:\/\/devblogs.microsoft.com\/oldnewthing\/wp-json\/wp\/v2\/posts\/105922\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/devblogs.microsoft.com\/oldnewthing\/wp-json\/wp\/v2\/media\/111744"}],"wp:attachment":[{"href":"https:\/\/devblogs.microsoft.com\/oldnewthing\/wp-json\/wp\/v2\/media?parent=105922"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/devblogs.microsoft.com\/oldnewthing\/wp-json\/wp\/v2\/categories?post=105922"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/devblogs.microsoft.com\/oldnewthing\/wp-json\/wp\/v2\/tags?post=105922"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}